{ "music": { "title": { "src": "music/title.ogg", "ms": 54340, "loop": true, "bpm": 106, "key": "E2 dorian", "notes": "Title rotation. Funk bed with talk-sung drifter vocals and dark-humor wasteland lyrics — a Karplus-Strong string through a modelled amp and 4x12 cabinet.", "bytes": 595417, "alts": [ "ride_or_rust", "truck_nasty", "truck_nasty_slow", "hammer_down", "derby_cut", "death_by_diesel", "get_nasty", "sizzle_on_the_gnar", "chrome_dome" ] }, "drive": { "src": "music/drive.ogg", "ms": 36923, "loop": true, "bpm": 104, "key": "E2 dorian", "notes": "Main driving loop. Dorian's major-6 is grit WITH forward momentum.", "alts": [ "warpath", "strap_in", "area_88", "acid_storm", "gnaw_what_i_mean", "hard_in_the_chamber", "truckin_sick", "mutant_on_the_one", "dirty_drift", "bodak_gallop", "rust_in_trap", "free_for_all", "cat_scratch_chamber", "thirty_six_hairs", "aqua_net_bodak", "one_nation_blast", "three_dollar_chamber", "hands_on_the_platter", "ride_the_fader", "four_steps_to_the_bloom", "shellac_and_subwoofer", "all_attack_no_sustain", "eight_drunken_immortals", "chrome_relic_minigun", "junk_gang", "scrap_baron2", "nasty_on_delayed_arrival", "dont_touch_the_trim", "chromebeard", "super_haul_haul", "boost_gauge", "deadhead", "who_clipped_me", "what_clipped_me", "son_of_diesel_money", "greenhouse" ], "bytes": 446079 }, "rampage": { "src": "music/rampage.ogg", "ms": 29091, "loop": true, "bpm": 132, "key": "E2 minor_pent", "notes": "Multiplier >= 5. Tight and loud — reverb pulled back so it hits hard.", "alts": [ "necro_bloom", "down_for_robotics" ], "bytes": 363511 }, "boss": { "src": "music/boss.ogg", "ms": 43636, "loop": true, "bpm": 88, "key": "C2 locrian", "notes": "The Undertaker duel. Locrian's flat-5 never resolves — permanent dread.", "bytes": 460565, "alts": [ "scrap_baron" ] }, "shop": { "src": "music/shop.ogg", "ms": 40000, "loop": true, "bpm": 72, "key": "G2 dorian", "notes": "Shop/safe. Currently the game KILLS music here (game.js:3365) — give it something warm instead of silence.", "alts": [ "temple_of_boom", "undergrind", "ro_schizzle", "far_horizon", "ocarina_flats", "cantina_dust", "something_coming", "dust_and_diesel", "cold_start", "radio_kept", "tape_rot", "night_haul", "wig_tight" ], "bytes": 468548 }, "gameover": { "src": "music/gameover.ogg", "ms": 38400, "loop": true, "bpm": 50, "key": "D2 aeolian", "notes": "Death. Slowest, darkest, longest tail.", "bytes": 433212 }, "warpath": { "src": "music/warpath.ogg", "ms": 123243, "loop": true, "bpm": 148, "key": "E2 minor_pent", "notes": "Drive alt. Classic-metal warpath riff, transposed mid-section, big outro.", "bytes": 1872374 }, "strap_in": { "src": "music/strap_in.ogg", "ms": 129231, "loop": true, "bpm": 104, "key": "E2 dorian", "notes": "Drive alt. Swagger-rock engine-warmup; flange on the chug.", "bytes": 1779686 }, "area_88": { "src": "music/area_88.ogg", "ms": 101053, "loop": true, "bpm": 152, "key": "E2 dorian", "notes": "Drive alt. UN Squadron homage - melodic lead over a running rhythm section.", "bytes": 1527776 }, "acid_storm": { "src": "music/acid_storm.ogg", "ms": 125217, "loop": true, "bpm": 92, "key": "E2 minor_pent", "notes": "Drive alt. Psychedelic distortion, death-metal leadups, the gnarly trippy solo. Wah + flange.", "bytes": 1809835 }, "necro_bloom": { "src": "music/necro_bloom.ogg", "ms": 103636, "loop": true, "bpm": 176, "key": "E2 phrygian", "notes": "Rampage alt. Blast-tempo psychedelic metal - the multiplier deserves 176 bpm.", "bytes": 1613156 }, "temple_of_boom": { "src": "music/temple_of_boom.ogg", "ms": 120000, "loop": true, "bpm": 96, "key": "E2 minor_pent", "notes": "Shop alt. 90s boombap, deep psychedelic echoes - count your scrap to it.", "bytes": 1693609 }, "undergrind": { "src": "music/undergrind.ogg", "ms": 76800, "loop": true, "bpm": 112, "key": "E2 dungeon", "notes": "Shop alt. Mario-dungeons homage from the sparkone-music _dungeon layer.", "bytes": 999924 }, "ride_or_rust": { "src": "music/ride_or_rust.ogg", "ms": 67200, "loop": true, "bpm": 200, "key": "E2 phrygian", "notes": "Title rotation. Thrash with a renaissance CANON inside it: one subject in four voices along the harmonic series of E, over blast beats, resolving onto a held note whose overtones cascade.", "bytes": 864428 }, "truck_nasty": { "src": "music/truck_nasty.ogg", "ms": 73846, "loop": true, "bpm": 208, "key": "E phrygian dominant", "notes": "Title rotation. Three lead guitars in real counterpoint (melody, diatonic third, and a third above that), a 40-hit-per-second gravity blast, flanger-and-wah psychedelia, gang vocals.", "bytes": 1015846 }, "truck_nasty_slow": { "src": "music/truck_nasty_slow.ogg", "ms": 58182, "loop": true, "bpm": 132, "key": "F# minor", "notes": "Title rotation. The same world at half speed — 808s, trap hats, one six-note motif and a drawled eight-bar verse. 613 events where the thrash cut has 1,733.", "bytes": 810736 }, "hammer_down": { "src": "music/hammer_down.ogg", "ms": 109714, "loop": true, "bpm": 140, "key": "F#1 phrygian dominant", "notes": "Title rotation. The long one: a SUBDIVISION LADDER at fixed tempo — quarters, eighths, eighth-triplets, sixteenths, gallop, sextuplets — then a half-time drop and back to quarters for the fold. Bass is a synthesised 808 matched to a MEASURED TR-808 (51.8 Hz, 98% under 60 Hz); the reference samples are GPL and were deliberately not used. Instrumental.", "bytes": 1307058 }, "derby_cut": { "src": "music/derby_cut.ogg", "ms": 82285, "loop": true, "bpm": 140, "key": "G1 minor", "notes": "Title rotation. THE SAMPLED ONE — every drum is a real recording from Hollow Sun's public-domain Classic-808 kit, not a synthesised model of one, and the 808 kick is pitched to play the bass line. Three Southern women trade 23 takes of demolition-derby double entendre over it. Melody is the 808 cowbell and congas pitched to notes, so there is no lead synth anywhere.", "bytes": 1013695, "credit": "Classic-808 — Hollow Sun (Steve Howell), Public Domain" }, "death_by_diesel": { "src": "music/death_by_diesel.ogg", "ms": 121263, "loop": true, "bpm": 190, "key": "E/G/A phrygian", "notes": "Title rotation. REDNECK DEATH METAL on a REAL ACOUSTIC KIT — AVL's Red Zeppelin 4pc, 118 samples with five recorded velocity layers per drum, so a blast beat and an accent are different recordings rather than the same one louder. Melody is a real recorded guitar re-amped through the modelled cabinet; the high-gain rhythm guitars stay synthesised because a sample pitched to E2 loses its top end. 808 accents sit under the riff at 40-60 Hz. Progressive by construction: three key centres (E->G->A), an accent pattern of SEVENS over an unbroken 4/4 grid, every return transformed. The mutant barks five times, only at section seams. FUNK BREAKDOWN at bars 33-48 (2026-08-16): the 16 bars bracketed by \"'Merica! Truck yeah!\" and \"Hell yeah, brother.\" were sevens-over-4/4 palm-muted chug and are now half-time funk — chicken-scratch sixteenths with holes, slap bass, Dazz-style stabs, clav in the second half — under the orchestra's own scheduled flanger+auto-wah bus (instr 91) at 1.9 Hz, one wah cycle per half-bar of the half-time pocket.", "bytes": 1572205, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Seagul Acoustic Git — unattributed; shipped with the Csound documentation since 1999 (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "get_nasty": { "src": "music/get_nasty.ogg", "ms": 111628, "loop": true, "bpm": 172, "key": "F#1 phrygian dominant", "notes": "Title rotation. A DEATH-METAL PASSACAGLIA WITH AN 808. The form is the idea: a passacaglia is a repeating bass line with continuous variation above it, and an 808 in modern rap is already structurally load-bearing in exactly that way — so the brief's two halves are joined by a form rather than by layering. The ground is the chromatic lament bass, F#1 descending by semitone to the dominant (F# F# F E D# D C# C#), and all ten cycles transform it audibly: plain, TRANSPOSED up a fourth for the hooks, INVERTED so the descent becomes an ascent under verse 2, in DIMINUTION at 4x speed under the solo, and in AUGMENTATION at half speed through the coda. F# is chosen BY THE SAMPLE — Classic-808's long kick measures 49.8 Hz, so the tonic sits one semitone from the recording's own pitch and the hooks, transposed up a fourth, land on the best-sounding version of the 808 in the whole track. Two felt tempos on one unbroken grid: the kit plays 172 while the ground and the drawled verses sit in half-time at 86, which is what lets a blast beat and a Southern drawl share a track without a tempo change. FIRST TRACK IN THE FLEET WITH A REAL DISTORTED ELECTRIC GUITAR — the public-domain GM bank's DistortionGuitar preset, re-amped through the modelled cabinet; a real picked bass holds 120-300 Hz where the 808 cannot reach and a phone can; and the acoustic kit is AVL's Black Pearl 4pc rather than Red Zeppelin, chosen because at 172 bpm a blast beat on a roomy kit is porridge. Toms are addressed by measured role, not by General MIDI number. Three desert women trade 18 takes about trucks getting nasty over it, screamed in a three-voice gang stack on every hook; the arrangement removes the mid band entirely for bars 41-48 so a deadpan damage report has a hole to land in.", "bytes": 1448730, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain); GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)", "license": "CC-BY-SA-3.0" }, "gnaw_what_i_mean": { "src": "music/gnaw_what_i_mean.ogg", "ms": 125217, "loop": true, "bpm": 92, "key": "A1 phrygian", "notes": "Drive rotation. DEATH METAL x WU-TANG, OVERDOSED ON 808. The two genres are joined by a key and a tempo rather than by alternating between them. A PHRYGIAN because its flat-2 is the one interval both idioms are built on — RZA's dusty menace and death metal's home dissonance are the same note — so the 808 leans on the B-flat in bar 2 of every phrase and the drop-A chugs answer it, and nothing has to modulate. 92 BPM because it is simultaneously the boom-bap pocket (swung 16ths, and the swing is applied to off-8ths only so the offbeats read late instead of the bar reading slow) and, at straight 16ths on the SAME grid, a 368-hit-per-minute blast beat: the drums never change tempo, they change SUBDIVISION, and the 808 underneath does not move at all. That continuity of low end through both idioms is why it reads as one record. The 808 is the only bass instrument on the track — instr 25 SL808, 88 notes of which 23 glide — so the same voice is the hip-hop bass under the break and the metal low end under the blast, and it never hands over. Kit is AVL's Black Pearl rather than Red Zeppelin because at 16ths a roomy 4pc is still sounding under the next four hits. Two spoken hooks from two stoned, unhurried women, one instance each: the arrangement VACATES bars 12 and 24 for them and the mix ducks only 300-4000 Hz, so the low end runs continuously underneath. Both takes were pitch-corrected to a MEASURED target (187.5->158 Hz and 205.1->166 Hz) with rubberband, formants deliberately NOT preserved because letting them fall with the pitch models a larger vocal tract, which is most of what reads as mature and close. Gangster rhythm solo over the boom-bap groove at bars 29-40, not over a metal section — the groove is the co-star. Bars 41-48 play both drum idioms at once, the only place they meet, so the fusion is the conclusion rather than a stunt.", "bytes": 2956009, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "sizzle_on_the_gnar": { "src": "music/sizzle_on_the_gnar.ogg", "ms": 152727, "loop": true, "bpm": 88, "key": "E1 phrygian dominant", "notes": "Title rotation. NECRO_BLOOM x CYPRESS HILL, with a bassline built for a car. The cross is made with a tempo and a key that are both INHERITED rather than invented: 88 bpm is the boom-bap pocket and 88 x 2 = 176 is exactly necro_bloom's tempo, so the blast sections are that track's own engine running on the same grid the groove sits in — nothing modulates and nothing changes tempo, the subdivision doubles. E PHRYGIAN DOMINANT is necro_bloom's own mode and root; its major third stacked on a flat second is the gnarly colour the parent is named for, and the flat second is the note the boom-bap idiom already leans on. TWO KITS FROM THE RACK, which is what the rack exists for: AVL Red Zeppelin (roomy, fat, cracked backbeat with a rimshot a 16th late) for the Cypress sections, AVL Black Pearl (tight, fast decay) for the blast, because at 176 a roomy 4pc is a wash. THE BASSLINE IS AN ENGINEERING SPEC AND IT CONTRADICTS THIS FLEET'S USUAL ONE: gnaw_what_i_mean drove its 808 hard on purpose because a 55 Hz fundamental is inaudible on a phone and the saturator's harmonics are what let the ear infer it, but a subwoofer reproduces 30-60 Hz for real and gets cabin gain on top, so here the fundamental is present instead of implied. Root E1 = 41.2 Hz; the sub is MONO by construction (instr 2 ends `outs aOut, aOut`) and measures 0.9995 L/R correlation below 120 Hz, because most car installs sum to one sub and stereo content down there partially cancels; no two bass notes ever overlap, asserted at build time, because two 41 Hz notes sum by PHASE and an overlapping line has an unpredictable level in a vehicle nobody can audition here; and the bass runs dry (reverb send 0.02) because reverb at 41 Hz is the definition of boom. Measured on the shipped file: 30-60 Hz sits 1.4 dB under full mix level and the 100-250 Hz mud band 16.9 dB under it. The delay bus is the house dotted-eighth ping-pong and it is treated as an instrument rather than a coating — the nasal squeal ostinato and the dark chord stabs are mostly TAIL. The rhythm solo runs over the boom-bap groove, not the blast, alternating held wah-swept notes against tight 16th cells under a flanger the arrangement automates per section (instr 91, scheduled). Two hooks from one nine-foot baritone: the takes came back at 79.9 and 83.9 Hz median F0 and were corrected to a common 68 Hz — just under a trained bass singer's floor — with formants deliberately NOT preserved, because a bigger resonator is the only thing 'nine feet tall' can mean acoustically. The duck splits at 120 Hz rather than 300: everything the subwoofer is for runs continuously, and the band where his body and every formant live opens for him. He lands 7.6 and 8.9 dB over the bed in the consonant band.", "bytes": 3469702, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "hard_in_the_chamber": { "src": "music/hard_in_the_chamber.ogg", "ms": 109714, "loop": true, "bpm": 140, "key": "G1 phrygian", "notes": "Drive rotation. HARDCORE TRAP THAT MODULATES INTO BOOM BAP ON A DOTTED QUARTER. Instrumental. Denzel Curry and Waka Flocka at 140 on one side, RZA at 93.333 on the other, and the join is a METRIC MODULATION rather than an edit: 140 and 93.333 are 3:2, so the boom-bap QUARTER NOTE is the trap's DOTTED QUARTER and the track never actually changes speed — you start counting it differently. That only works if every seam lands on a beat of both grids, so it is asserted at import instead of trusted: 42 trap bars = 112 dotted quarters exactly, 12 boom bars = 18 trap bars exactly, and the whole thing is 64 bars of one 140 bpm pulse. Verified from the rendered audio by autocorrelating spectral flux per section — 139.98/139.98/139.99/140.02 across the four trap sections and 93.31/93.35 across the two boom-bap ones. Bars 39-42 TELEGRAPH it with four bars of 3-against-4 on the cowbell and the 808, and those bars deliberately start 3.4 dB DOWN, because a polyrhythm underneath a section as dense as the one before it is not audible as a polyrhythm. FOUR BASS VOICES WITH FOUR JOBS, which is what keeps an 808 overdose heavy instead of muddy: instr 25 SL808 carries the line and glides (with an octave layer at 98 Hz, because the G1 tonic at 49 Hz is not reproduced by a laptop at all); instr 19 holds the floor between hits, since a sample decays and a synth does not; a sine sits under section seams; and a detuned-saw Reese owns 80-220 Hz, which is where 'heavy' actually lives on a phone. G PHRYGIAN CHOSEN BY THE SAMPLE — Classic-808's Kick_long measures 49.8 Hz, so G1 sits three tenths of a semitone from the recording's own pitch and the tonic is the best-sounding note the instrument has. Phrygian because its flat-2 is the interval both idioms are built on: the A-flat over G is the Lex Luger horror lean and RZA's dissonance, the same note, so nothing has to modulate harmonically when the tempo does. Cardi-style RHYTHM SOLO at bars 29-38 — a rhythm part written rhythm-first and pitched second, 11 of its 17 sixteenths landing off the beat, crowding the front of the bar so the 808 can answer into the hole at the back. Mastered to -12.6 LUFS ON PURPOSE, to sit inside this rotation rather than under it (QUAL-319).", "bytes": 1335607, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "truckin_sick": { "src": "music/truckin_sick.ogg", "ms": 121379, "loop": true, "bpm": 87, "key": "G1 aeolian (tritone)", "notes": "Drive rotation. AESOP ROCK MURK CROSSED WITH DENZEL CURRY AGGRESSION, TWO 808s, AND A DOUBLE-BASS DRILL. THE BEAT DISAGREES WITH ITSELF BY CONSTRUCTION: the 808 ground is FIVE bars long and the drum pattern is FOUR, so for twenty bars the downbeat means something different every time round and the loop never quite repeats — which is the Aesop half done as arithmetic rather than as vibe. LCM(5,4) = 20, so the two cycles agree exactly once, at bar 25, AND THAT IS THE BAR THE DOUBLE BASS LANDS ON. The alignment is the drop; nothing else marks it. THE DOUBLE 808 IS SEPARATED BY ENVELOPE, NOT BY PITCH, and that came out of measuring the pack rather than assuming: Classic-808 ships two kicks and BOTH fundamental at 49.80 Hz, so there is no pitch difference available to tell them apart with. What differs is decay — 2.23 s against 0.27 s. Kick_long therefore plays the gliding ground and Kick_short plays sixteenths and thirty-second doubles, and the two stack on the same note without mud because they never occupy the same part of the envelope. 138 sustain notes and 512 punch notes. A picked upright doubles the ground two octaves up at 120-300 Hz, which is the only bass band a phone speaker actually passes. G AEOLIAN LEANING ON THE TRITONE: G1 is 49.0 Hz, 0.28 semitones from what both kicks measure, and the flat-5 (D-flat) is simultaneously the horror interval a dusty minor loop is built on and the entire vocabulary of a Megadeth riff — so the murk and the metal argue over the same note and the drill arrives without a key change. TWO KITS BY MEASUREMENT: Red Zeppelin (roomy) plays the boom-bap pocket where the space between hits is the sound; Black Pearl (tight) plays the drill, because the double bass runs thirty-seconds at 87 bpm — 11.6 hits a second — and a roomy 4pc at that rate is porridge. The lead solo is NINE NOTES ACROSS EIGHT BARS, sparse on purpose, every long one bent. Four robot ad-libs (\"Sup.\", \"Word.\", \"Truckin' sick.\", \"Word.\") synthesised by eSpeak NG rather than by the neural TTS the rest of the fleet uses — a formant synthesiser has no micro-prosody to remove, which is exactly what makes a processed voice read as a machine instead of as a man doing an impression. All three takes pitch-corrected to ONE measured target (58.27 Hz, the minor third) because eSpeak's own spread was 7.5 semitones and a robot whose pitch wanders is a person; formants deliberately NOT preserved, so the vocal tract grows with the drop. The \"fan\" is 13 Hz amplitude modulation, measured at +12.8 to +20.4 dB above its neighbourhood. Mastered to -12.6 LUFS on purpose, to sit inside this rotation rather than under it (QUAL-319).", "bytes": 1502700, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "mutant_on_the_one": { "src": "music/mutant_on_the_one.ogg", "ms": 117551, "loop": true, "bpm": 98, "key": "A1 dorian", "notes": "Drive rotation. LIMP BIZKIT CROSSED WITH PARLIAMENT-FUNKADELIC, AND THE METAL RIFF AND THE FUNK LICK ARE THE SAME RHYTHM — not similar, the SAME sixteen-slot tuple through the same generator, played first as a drop-tuned palm-muted chug and later as a clavinet lick with chicken-scratch guitar on it. Nu-metal took its groove from funk (syncopated sixteenths landing hard on the ONE, which is George Clinton's whole doctrine), and the track argues that by playing it rather than stating it: bars 39-44 run the chug and the clav together, which is only possible because it was one rhythm the entire time. Verified from the rendered audio, band-limited to 700-3500 Hz where both instruments live: the metal section correlates +0.881 with the score's riff and the both-at-once section +0.905. A DORIAN because the major sixth over a minor tonic is the funk interval every P-Funk vamp sits on, and dorian is also where drop-tuned nu-metal lives — one mode, both idioms, no modulation anywhere. THE BASS WAS MEASURED, NOT CHOSEN: all 26 bass-capable voices in the orchestra were rendered playing the same note, and the result was that EVERY SYNTHESISED BASS VOICE HAS A 40+ dB HOLE AT 60-120 Hz. That is arithmetic rather than a bug — they all saturate with tanh, tanh is an ODD function, so it generates 3f and 5f and never 2f, which at a 55 Hz tonic leaves 110 Hz empty. 110 Hz is the octave a laptop and a phone reproduce best of anything. So the bass here is four voices each measured best at one job and none duplicating another: instr 2 SUB is the floor (best 20-60 Hz measured, and the longest sustain at -2.0 dB over 300-600 ms); instr 25 SL808 is the character and owns 120-250 Hz through its odd harmonics; GM Synth Bass 3 fills the hole at -12.8 dB against the 808's -56.0, because a SAMPLED bass has even harmonics and no amount of drive on an 808 can manufacture them; and GM Slap Bass 2 is the transient at 2.7 ms to peak, the fastest of the twenty-six, which also happens to be the actual Parliament instrument so it plays the breakdown. Drums: 808 kit for the nu-metal half with thirty-second-note rat-a-tat snare rolls, a SEEDED chaos section at bars 19-22 (dense, erratic, and byte-identical on every render because the RNG is seeded and the orchestra never calls Csound's `seed`), and a real acoustic kit for the funk. The breakdown runs under the orchestra's own scheduled auto-wah (instr 91). Three hooks, three separate mix chains: a shout with presence and a 90 ms slapback, a mouth-percussion line treated as a drum fill with no echo at all, and a four-second soul swell with low-mid warmth and a long tail over two bars the arrangement empties completely for it. Mastered to -12.6 LUFS on purpose, to sit inside this rotation rather than under or over it (QUAL-319).", "bytes": 1507369, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "dirty_drift": { "src": "music/dirty_drift.ogg", "ms": 116044, "loop": true, "bpm": 91, "key": "F#1 aeolian", "notes": "ALBUM I, TRACK 3. Boom bap with the measured 808 stack, in the Ol' Dirty Bastard register, with death-metal ramps welded onto every transition. THE LOOP IS RIGID AND EVERYTHING OVER IT DRIFTS: what makes that register sound like itself is not being off — anyone can be off — it is being off AGAINST SOMETHING MECHANICAL. So the drums and the 808 are quantised dead to the grid and never move a sample, while the dusty melodic loop is displaced by up to 38 ms on a shape that runs EARLY through the first half of every four-bar phrase and LATE through the second, resolving to exactly zero on the phrase downbeat — rushing and dragging, not randomness. 38 ms is 23% of a sixteenth at 91 bpm; past about a third of a subdivision it stops reading as feel and starts reading as a mistake. Its BELL layer also runs 18 cents FLAT, so the two halves of the loop are out of tune with each other and beat slowly, the way a warped record does. That detune sits on the bell rather than the whole loop for a measured reason: instr 23 is key-mapped and Csound's sfplay3 TRUNCATES its note number — MIDI 53.82 and 53.00 render identically — so a soundfont preset cannot be detuned by cents at all, while the Hz-addressed bell can, verified at -18.0 cents on every partial. EVERY EIGHT-BAR SEAM IS A DEATH-METAL RAMP: bars 8, 16, 24, 32 and 40 drop the boom bap entirely for a SUBDIVISION ACCELERANDO — eighths on beat 1, sixteenths on 2, sextuplets on 3, thirty-seconds on 4, under tremolo-picked guitar — so the music accelerates while the tempo does not, and the next section lands exactly on the grid. Bass is the stack derived from measuring all 26 bass-capable voices in the orchestra: instr 2 SUB on the floor, instr 25 SL808 for character and the 120-250 Hz odd harmonics, and a SAMPLED bass filling the 60-120 Hz hole that tanh saturation provably cannot, because tanh is an odd function and makes 3f and 5f but never 2f. Three ad-libs in the register, original words throughout.", "bytes": 1521543, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "bodak_gallop": { "src": "music/bodak_gallop.ogg", "ms": 102400, "loop": true, "bpm": 150, "key": "G#1 harmonic minor", "notes": "ALBUM I, TRACK 4. Cardi B x Megadeth. THE RAP CADENCE IS THE METAL GALLOP WITH NOTES TAKEN OUT — not 'similar to', a strict subset, and the composer asserts it at import: every hit in the trap cadence lands on a sixteenth the thrash gallop already plays. GALLOP is twelve hits a bar with no holes anywhere; BODAK is eight, crowding the front of the bar and leaving the back open. Same figure, two densities, which is why bars 29-40 can lay the gallop guitar over the trap drums with the cadence on top and bend neither to fit. The arrangement is the proof: the cell is heard alone on a clean guitar first, then as full thrash, then with the holes cut in, then both at once. G# HARMONIC MINOR, chosen by the sample again — G#1 is 51.91 Hz against Classic-808's measured 49.80, the smallest stretch of any key on these records, and pitching slightly UP shortens the sample, which is what 150 bpm wants. Harmonic minor because its augmented second is the interval Megadeth built a career on and because trap has lived there for a decade: one scale, both idioms, no modulation. The solo is HARMONISED THIRDS rather than speed — that is the actual signature being borrowed; every note is played twice, the line and the line a scale-third above it. Bass is the measured four-voice stack. Three voices: a boast delivered flat and certain rather than shouted, a gang shout for the metal half, and DROID — the flat formant-synthesised machine from ALBUM I's 'Truckin' Sick' — answering the boast with a single unimpressed 'Word.' It is the SAME RECORDING, pitch-matched to the same 58.27 Hz target the other record pinned it to, because a callback that is merely similar is an impression.", "bytes": 1375431, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "rust_in_trap": { "src": "music/rust_in_trap.ogg", "ms": 108000, "loop": true, "bpm": 160, "key": "B1 aeolian (chromatic descent)", "notes": "ALBUM II, TRACK 2. Megadeth x Denzel Curry. BOTH IDIOMS ARE BUILT ON A FALLING LINE AND THEY DISAGREE ONLY ABOUT WHETHER PITCH IS CONTINUOUS. Mustaine's writing signature is the chromatic descent — a riff walking down in semitones, each step a separate attack. Trap's is the 808 glide — a bass sliding between notes so the space between them is filled in. Those are the same gesture at two resolutions of pitch, so this record states one six-note descent (B1 down to F#1) twice: the guitar plays it as six decisions, the 808 plays it as five glides, and bars 35-48 run both at once, which is the only place the ear can hear they are the same line. THE KEY WAS CHOSEN BY WHERE THE DESCENT LANDS, not where it starts: B1 is 61.7 Hz and six semitones down is 46.2, so the whole chromatic run stays inside the band Classic-808's measured 49.80 Hz kick sounds best across — a descent starting lower would walk out of the instrument by the fourth step and the last two notes would be flab. Half-time trap clap on 3 against a thrash backbeat on 2 and 4, same grid, twice the felt speed. The solo is harmonised thirds rather than speed. Three voices: a deadpan title pun delivered as though it were a weather report, a gang chant, and DROID — the flat machine from Album I — turning up on the metal record to say 'Truckin sick.' with no detectable opinion. Same recording as Album I, pitch-matched to the same target: a callback that is merely similar is an impression.", "bytes": 1427608, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "free_for_all": { "src": "music/free_for_all.ogg", "ms": 114930, "loop": true, "bpm": 142, "key": "E1 aeolian (one chord)", "notes": "ALBUM I, TRACK 5. Waka Flocka x Ted Nugent. NEITHER OF THESE IDIOMS EVER CHANGES CHORD, SO THIS RECORD DOESN'T EITHER. A Lex Luger trap beat is one minor loop for four minutes; 'Stranglehold' is one vamp for eight. That is the same decision in both cases — that ENERGY IS THE STRUCTURE and harmony is not going to help — so the bass here plays E and only E for a hundred and fifteen seconds, and the composer asserts it at render rather than claiming it: the pitch is recovered back out of instr 25's playback ratio and every one of the 136 808 notes is checked. Everything that happens instead happens in REGISTER, DENSITY and TEXTURE, and that is measurable: from the opening vamp to the peak, broadband level barely moves (which is what a constant 808 predicts) while 0.5-2 kHz climbs 19 dB and 2-6 kHz climbs 25 dB. The Lex Luger orchestral stab and the live-wire guitar hit play the SAME seven-hit figure — brass first, then guitar, then trading it bar by bar — because a stab is a stab and the rhythm does not care which section is holding it. E MINOR WITH THE TONIC BELOW THE SAMPLE: E1 is 41.20 Hz against Classic-808's measured 49.80, so the 808 is pitched DOWN 3.3 semitones, and loscil3 resamples, so pitching down makes the sample LONGER. This is the only track on either record that wants that — a Lex Luger 808 is supposed to still be ringing when the next one lands. Every other track here pitches up for tightness. The solo is pinch harmonics and bends that overshoot: instr 17 reads a NEGATIVE slide as a pinch, which is most of the vocabulary being borrowed. Three voices, including one deadpan observation about the record it is standing on, and DROID's flat 'Sup.' carried over from Album I.", "bytes": 1416211, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "cat_scratch_chamber": { "src": "music/cat_scratch_chamber.ogg", "ms": 115200, "loop": true, "bpm": 100, "key": "A#1 aeolian (guitar +22c, loop +42ms)", "notes": "ALBUM II, TRACK 3. Ted Nugent x ODB. TWO KINDS OF WRONG, POINTING IN OPPOSITE DIRECTIONS. These are the two least disciplined things in the corpus and they work together because they are undisciplined about DIFFERENT AXES: the guitar is wrong in PITCH, sitting 22 cents sharp for the entire record — Nugent's habit is a bend that overshoots and never quite comes back — while the loop is wrong in TIME, dragging 42 ms behind, which is 28% of a sixteenth. The rhythm section is dead on the grid and dead in tune, both, all the way through, and is the only reference either error is measured against. Neither is a wobble that resolves: the previous record on this shelf (Dirty Drift) oscillated its drift and returned to zero every four bars, and this one uses CONSTANT offsets instead, which is a stranger and more unsettling thing — nothing ever comes back into agreement and the ear stops waiting for it to. The sharp is only possible on Hz-addressed voices; instr 23 is key-mapped and sfplay3 truncates its note number, so a soundfont preset cannot be detuned by cents at all. Tonic B-flat1 at 58.27 Hz, which is exactly the frequency every DROID take on these records is pinned to — a coincidence worth keeping.", "bytes": 1494832, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "thirty_six_hairs": { "src": "music/thirty_six_hairs.ogg", "ms": 120000, "loop": true, "bpm": 96, "key": "D#1 aeolian (two rooms)", "notes": "ALBUM I, TRACK 6. Wu-Tang x hairband. THE PARTS NEVER CHANGE. THE ROOM DOES. What separates 1993 Staten Island from 1987 Sunset Strip is not the notes and not the tempo — it is REVERB and STEREO WIDTH. A RZA loop is dry, narrow and close, as though recorded in a cupboard; a hairband record is enormous and washed. So this plays the same figure, the same riff and the same drums throughout and moves only those two knobs, and both are measurable off the finished master: the chamber sections sit at -29.4 dB side-to-mid and the arena at -21.4 dB, eight decibels wider, with the reverb send ten times higher. The two chamber sections either side of the arena measure within 0.1 dB of each other, which is the proof that nothing was rewritten. Bars 41-44 alternate the room bar by bar so the switch itself becomes the rhythm. Tonic E-flat1 at 38.89 Hz, deliberately the lowest on either record and 4.3 semitones under the sample — loscil3 resamples, so pitching down lengthens it, and a subterranean 808 is the chamber half of the brief while the arena gets its size from the room instead.", "bytes": 1580119, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "aqua_net_bodak": { "src": "music/aqua_net_bodak.ogg", "ms": 112500, "loop": true, "bpm": 128, "key": "F1 aeolian (the hook is the hole)", "notes": "ALBUM II, TRACK 4. Hairband x Cardi B. THE HOOK IS THE HOLE. The Bodak-shaped cadence crowds the FRONT of the bar and leaves the back of it completely empty — that imbalance is the whole character of the flow. Glam metal's hook is a gang shout on the backbeat. Put them in the same bar and the shout lands exactly in the space the cadence vacated: the two parts are not competing for room, one of them IS the other one's silence. The composer asserts it — all six cadence hits fall inside the first nine sixteenths, slots 9-15 are empty, and the hook is placed on slot 12 — and refuses to build if a later edit puts a note in the back half of the bar, because then the hook has nowhere to land and the record stops making its point. Tonic F1 at 43.65 Hz, 2.3 semitones under the sample so the 808 rings a little long, which is what a hook needs to sit on top of.", "bytes": 1539577, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "one_nation_blast": { "src": "music/one_nation_blast.ogg", "ms": 115385, "loop": true, "bpm": 104, "key": "C2 aeolian (one grid, two dynamics)", "notes": "ALBUM II, TRACK 5. Death metal x Parliament-Funkadelic. A BLAST BEAT AND A FUNK SIXTEENTH ARE THE SAME NOTES; THE DIFFERENCE IS DYNAMICS. Both play every sixteenth of the bar. What funk does — and it is George Clinton's entire doctrine — is put the accent on THE ONE and bury everything between it in ghost notes, so the bar has a shape. A blast beat plays the identical grid with no shape at all: flat, level and undifferentiated, which is the point of it. So this uses ONE onset array and TWO velocity arrays, asserted at import: the funk array has a variance of 6.34 with its maximum on slot zero, the blast array has a variance of exactly zero, and nothing is added or removed between them. Take the accent away and the groove becomes a blast; put it back and the blast becomes a groove. Bars 29-40 alternate bar by bar so the switch is the rhythm, and bars 41-46 are the only place a funk bass and blast drums coexist. Tonic C2 at 65.41 Hz — 4.7 semitones ABOVE the sample, the largest upward stretch on either record, because at 416 sixteenths a minute a long 808 is a smear.", "bytes": 1510059, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "three_dollar_chamber": { "src": "music/three_dollar_chamber.ogg", "ms": 114783, "loop": true, "bpm": 92, "key": "G1 aeolian (the subtraction)", "notes": "ALBUM II, TRACK 6. THE CLOSER. Limp Bizkit x Wu-Tang x Aesop Rock. THE SUBTRACTION. Every other track on these two records makes its argument by ADDING something — a second idiom, a second reading of one rhythm, a second kind of wrong. The closer makes it by taking things away: it opens with all three foundations playing at once and removes exactly one layer every six bars, in order, until the only thing left is the 808 — which is what all twelve tracks were built on and what the whole project has been about. Six layers, then five, four, three, two, one, and the composer asserts the count is monotonically decreasing, so a closer that quietly re-adds something in the last eight bars will not build. Every voice on it is a callback: the shout that opened Album II, DIRTY's cackle as his own layer is removed, and DROID saying 'Word.' for the fourth time across both records — the same recording every time, pitch-matched to the same 58.27 Hz. The last line on either album is the only one written for this track. Tonic G1 at 49.00 Hz: the key Album I opened in, 0.28 semitones from what the Classic-808 kicks actually measure, so the last thing anybody hears is the sample very nearly at its own pitch, alone, which is where all of this started.", "bytes": 1494164, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "hands_on_the_platter": { "src": "music/hands_on_the_platter.ogg", "ms": 112340, "loop": true, "bpm": 94, "key": "F#1 aeolian (the scratch is the cast)", "notes": "Old-school boom bap: Mobb Deep, Wu-Tang and Pharoahe Monch. THE SCRATCH IS THE CAST. There are no scratch samples on this machine — every allow-listed pack was searched and the only Hydrogen TR-808 kit present is GPL, which the licence gate refuses — so the crate was CUT rather than found, and the source material is the two albums' own vocal takes: DROID's 'Word.', DIRTY's cackle, the hype shout, the gang's 'IN THE GAP!' and SMOKE's boast, each dragged across the platter five different ways. A DJ's crate is other people's records; this one is ours, which is the only reason a track this dependent on scratching carries no licence question at all. THE SCRATCHES ARE SYNTHESISED RATHER THAN SAMPLED because a scratch is not a timbre — it is a function from output time to READ POSITION, and its derivative is the playback rate, which is the same physical resampling a record under a needle performs. Measured on the finished one-shots: the BABY runs 50% of its length backwards (rate -3.14 to +3.14), the TEAR 17%, and the TRANSFORMER never reverses at all (rate pinned at +0.85) because there it is the FADER doing the work and not the hand — eight chops with the pitch untouched. That distinction is the difference between the two commonest scratches there are, and it falls out of the path function rather than being imitated. Cubic interpolation throughout: at 3x resampling, linear is audibly gritty in the top octave. The routine escalates across the record and it was checked by rendering the bed twice with the scratches muted — they sit 24.0, 18.2 and 15.6 dB under the full mix across the three sections. Heavy stack underneath: 80-120 Hz is the strongest of any track in this collection at -6.6 dB relative. Rat-a-tat snares are three 808 snares inside one sixteenth, rising and panned across. F# minor, tonic 46.25 Hz.", "bytes": 1528743, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "ride_the_fader": { "src": "music/ride_the_fader.ogg", "ms": 117955, "loop": true, "bpm": 100, "key": "G1 dorian (96->118, 7-against-5-against-4)", "notes": "Scratching, 808 and funk bass, styled after Aesop Rock. TWO IDEAS ON TWO AXES. First, THE TEMPO IS A STAIRCASE AND EACH STEP IS A PITCH-FADER MOVE: a Technics fader is +/-8% and blending into a faster record means riding it, so the transitions ARE the fader — 96, then 100, 106, 112, 118, and a snap back to 96 for the loop. Nothing is quantised across a step; each section owns its own grid, computed from its own bpm, and the bar table is built once at import so no event can be placed on the wrong clock. Verified from the rendered audio by autocorrelating spectral flux per section: 99.97, 105.75, 112.28 and 117.88 bpm against 100, 106, 112 and 118 — every beat-bearing section within 0.25%. The total climb is 23%, well outside what a real pitch fader can do in one move, which is the joke. Second, THE PHRASES REFUSE TO AGREE, which is the Aesop Rock half and the same device 'Truckin' Sick' was built on, pointed here at the two elements the track is about: the drums run four bars, the SCRATCH ROUTINE runs SEVEN and the MURK runs FIVE, all counted absolutely and all ignoring the section boundaries. Pairwise coprime, so the three agree only every 140 bars against a 52-bar record — the same scratch lands on a different beat of a different bar at a different TEMPO every time it comes round, and nothing repeats exactly inside the track. The bass is written as funk rather than as root notes: the 808 takes the downbeat and the long notes, a slap bass plays the sixteenths between them INCLUDING the ghost notes, and neither ever plays the other's position — one part split across two instruments by register and envelope, which is the only way to get funk motion out of a bass that decays for two seconds. Every scratch is cut from this collection's own vocal takes. G dorian, tonic 49.00 Hz, essentially the 808 sample's own pitch.", "bytes": 1606113, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "four_steps_to_the_bloom": { "src": "music/four_steps_to_the_bloom.ogg", "ms": 119213, "loop": true, "bpm": 88, "key": "E1 phrygian (88->176, equal-tempered)", "notes": "404s and 808s, scratching, and a tempo ladder that ends on another record. THE TEMPO IS EQUAL-TEMPERED. 'Ride the Fader' stepped its tempo like a DJ riding a pitch fader — percentages chosen by feel. This one steps it like a KEYBOARD: four equal ratios of 2^(1/4), a quarter of a tempo-octave each, so 88 becomes 104.650, 124.451, 147.998 and finally 176.000 — and the top of the ladder is EXACTLY double the bottom, asserted at import. That last number is not chosen: 176 is necro_bloom's tempo, and the figure the record arrives at when it gets there was lifted from that track's own shipped score rather than approximated. Folding all 936 of its bell notes onto a one-bar sixteenth grid gives #####.#.#.+.#.+. — four consecutive sixteenths then a decaying alternation, with the tenth and fourteenth at half weight because they carry half the hits in the source. Its hi-hat was lifted the same way: every sixteenth, accented on the quarter. So the whole record is a ladder whose top rung is another record and whose last section is its own first section at double speed. THE DRUMS ARE 404s, and the character is in the SAMPLE rather than in the mix — because that is what a sampler is: you record into it, it keeps a degraded copy, and every trigger plays the degraded copy. Every one-shot went through a modelled SP-404 before the arrangement existed: band-limit in, sample-and-hold, 10-bit quantisation referenced to the signal's own peak, reconstruction filter out, vinyl sim. Measured on the finished one-shots, 99% bandwidth halves — a snare goes 11357 to 5607 Hz — and quantisation noise lands 44-54 dB under the signal, which is 10-bit territory. The reconstruction filter is the part that is easy to leave out and must not be: sample-and-hold CREATES high frequencies rather than removing them, and without a filter after it the box makes everything brighter and harsher, which is the opposite of the character. Scratches run on a five-bar cycle against four-bar drums — the Aesop Rock sprinkle, lighter than the three competing cycles of the previous record — and every one of them is cut from this collection's own vocal takes. E phrygian, necro_bloom's key, tonic 41.20 Hz.", "bytes": 1652742, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)" }, "shellac_and_subwoofer": { "src": "music/shellac_and_subwoofer.ogg", "ms": 117333, "loop": true, "bpm": 90, "key": "D1 aeolian (1928 mids + shaped 36.7 Hz sub)", "notes": "1920s public-domain chops over an asymmetrically-shaped 808. THE TWO HALVES OF THIS RECORD CANNOT OVERLAP, and not because they were EQ'd apart — because of physics and a century. Everything above 180 Hz was cut from four public-domain 78rpm recordings made between 1923 and 1928; everything below 120 Hz was synthesised. Shellac cannot hold deep bass: measured on these transfers, 99% of their energy sits below 2813 Hz and there is nothing under 60 Hz to begin with. Verified by rendering the bed twice with each half muted and differencing — the synthesised half owns 20-60 Hz by 58 dB and 60-120 by 20 dB; the 1928 half owns 300 Hz-1 kHz by 26 dB, 1-3 kHz by 37 dB and 3-8 kHz by 25 dB. They meet only at 120-300 Hz, which is where the handover was designed to happen and where the sampled phone-band bass sits. THE BASS IS ASYMMETRICALLY SHAPED, and that is why it survives a phone. The orchestra saturates with tanh, which is an ODD function — it makes 3f and 5f and never 2f — so at a 36.71 Hz tonic it puts energy at 110 and 183 Hz and leaves 73 Hz empty, which is the octave a phone speaker actually passes and the 40 dB hole every other track here patches with a layered sampled bass. An asymmetric shaper produces 2f directly: measured on the real Classic-808 kick, 60-120 Hz up 23.6 dB with 30-50 Hz UNCHANGED. It plays with instr 25's own drive at ZERO, deliberately — running it through tanh again would stack odd harmonics on top and bury the point. D MINOR, TONIC 36.71 Hz, CHOSEN FOR THE CAR RATHER THAN THE ROOM: below about 60 Hz a vehicle cabin behaves as a pressure vessel and gains roughly 12 dB/octave, which is why car audio moves trim a bookshelf speaker cannot. On a phone you hear the 73 Hz the shaper built; in a car you feel the 36.71 it left alone. Every sampled source is US public domain by statute (sound recordings first published before 1929, Music Modernization Act), with source URL, publication year, sha256 and fetch time recorded per file; the archive publishes no licence metadata for these items, so the basis is the date, and the date was parsed and checked rather than assumed. THE RECORD ARRIVES AND LEAVES RATHER THAN STARTING AND STOPPING. Drums, percussion and the 808 all read ONE build curve, and each element carries a threshold on it, so the kit assembles a limb at a time and the outro runs the same curve backwards to shed them in reverse. The ordering is the whole trick and the first cut had it inverted: gating the downbeat kick in first and the hats last measured, with the chops and bass muted, as -59.9 dB at bar 4 and -17.5 dB at bar 5 — silence to within 1.4 dB of the final level in one bar, then fourteen bars of creep. A kick carries so much more energy than a hat that the added layers were inaudible: a hard start wearing a build as a costume. With the loudest element gated in LAST the same measurement climbs 8.6 dB monotonically across all fourteen bars and the decay mirrors it at -8.7 dB, and the biggest single-bar jump in the whole intro falls from 13.6 dB to 6.7.", "bytes": 1472257, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "all_attack_no_sustain": { "src": "music/all_attack_no_sustain.ogg", "ms": 120000, "loop": true, "bpm": 88, "key": "E1 phrygian (real instruments, breaks at 176)", "notes": "Real instruments, an 80s car-audio palette, and four one-bar death-metal breaks. EVERY MELODIC AND HARMONIC VOICE ON THIS RECORD IS A RECORDING — Clavinet, Rhodes-type electric piano, percussive drawbar organ, a Stratocaster, fingered and slap electric bass, brass section, muted trumpet, baritone sax, vibraphone, the ORCH5-style orchestra hit, and a real distorted electric guitar multisampled at the six open-string pitches E1-A1-D2-G2-B2-E3. Nothing synthesised plays a note except one held sub-bass tonic and a noise bed. But auditioning all 28 candidates and MEASURING them is what actually shaped the arrangement: THE SUSTAINS ARE NOT REAL. That bank fits 488 samples into 4 MB by looping the steady part of every note, in loops as short as ten frames, and a ten-frame loop is perfectly periodic, so the recording's own noise floor is frozen out of it. Measured as the dB from the harmonic peaks down to the noise between them, the ATTACK of each instrument reads 20-47 dB — a real, noisy recording — and the SUSTAIN of the same note reads 46-99 dB, which is an oscillator. 27 of 28 instruments, median gap +37.7 dB. So the record is built ENTIRELY FROM ATTACKS: nothing melodic is held longer than 0.30 s, enforced by an assert that fails the render. That is a restriction and it is also exactly how 1980s funk-rap plays these instruments, which is why the constraint and the genre never pull against each other. THE ONE EXCEPTION IS THE POINT: a power chord IS sustain, so the distorted guitar in the break bars is deliberately held past the attack window and is the only voice on the record that is — checked in the opposite direction, that it is long AND that it only ever appears inside a break. THE BREAKS ARE THE TOP RUNG OF ANOTHER RECORD'S LADDER. 'Four Steps to the Bloom' climbed 88 to 176 BPM in four equal ratios of 2^(1/4); this record sits at the bottom rung and every break runs its drums on the top one, which is exactly double and is also necro_bloom's tempo — so a sixteenth there is a thirty-second here, asserted rather than trusted, at 32 kick drums a bar and 11.7 per second. The riff is E against Bb, the tritone Truckin' Sick is keyed on, and the scratches are cut from Truckin' Sick's OWN droid vocal takes, so both parents are in the structure rather than in a description. Measured against the flavor beat, a break bar carries +930 transients, +6.0 dB at 250-800 Hz and +9.8 dB at 1-4 kHz at the SAME overall level: it changes character, not volume. E PHRYGIAN, TONIC 41.20 Hz — phrygian for the flat second that is metal's whole identity, 41.20 Hz because it sits in the 30-45 Hz band a car cabin amplifies as a pressure vessel. TWO MEASUREMENTS CAUGHT ME OUT AND BOTH ARE IN THE CODE. The guitar was first written single-tracked and measured 17.6 dB under its own drums, so it is now double-tracked — two takes 7 ms apart, hard panned, which buys 6 dB honestly instead of turning one part up — and the whole melodic bus was contributing +0.1 dB overall until it was muted and re-rendered to prove it, which is the same complaint that started the record. It now owns +5.4 dB at 250-800 Hz. AND THE INTRO IS NOT A HOLE: the first master left bars 1-4 at -30 dB with bar 5 arriving +15.7 dB above them, which in a looping game track means every pass drops 18.7 dB into near-silence at the seam. The kit's own build gate fixes it without a second code path — at intro densities the kick is never reached, so those bars are hats, ghost snares and hand percussion. The drop is now +5.2 dB and the seam -7.6. FOUR THINGS WERE SUPERIMPOSED ON IT AFTERWARDS. A WAH: instr 23 gained an opt-in route into the orchestra's flanger/auto-wah bus, signalled by a NEGATIVE velocity because the p-field schema is full at p4..p10 — the same trick instr 17 already plays with a negative rate for a pinch harmonic — so the twelve melodic voices go through the pedal and the bass, guitar and drums stay dry. Isolated by differencing an A/B render, the pedal sweeps 867-2049 Hz and varies note-to-note brightness across 1064-2154 Hz. Its rate had to move: set at 1.45 Hz it sat on the beat (60/88 = 1.467 Hz), where a pedal is not heard as a pedal but as the pulse, and where it was unmeasurable because the centroid already wobbles at the beat with the bus switched off. MORE DISTORTION: instr 23 documents drive as 0..1 but nothing clamps it, and swept on this chord the 2-8 kHz content relative to 60-400 Hz goes +2.5 dB at 0.95 and +8.3 dB at 3.0 with `balance` holding the level — so the break amp runs at 3.0 and buys harmonics rather than volume. A break bar now measures +8.0 dB at 2-8 kHz against the flavor beat. A WHAMMY BAR, which had to be BAKED: sfplay3 truncates its note number, so a sampled bank cannot be bent by cents at all, and the bend is instead resampled into pre-rendered chords — a whammy is a function from time to pitch, resampling is a function from output time to read position, and the derivative of the second is the first, which is the same observation the scratches are built on. Verified against a synthetic tone: a warble asked for +-1.00 semitone measures +-1.01, a dive asked for -7.00 measures -6.63. AND CAR-STEREO DEMO DROPS on the last measure of each four-bar pattern — the bass-demo-record gesture, a tone that slides out of the musical register into pressure. The first attempt glided the 808 kick and failed for a reason worth keeping: a kick is a TRANSIENT, so it decayed before the glide was a third of the way down and measured as a step from 43 Hz to 21 Hz inside one analysis frame. The shipped version loops the 808's settled tail into a sustained tone first and sweeps that, 82 Hz down to 23 Hz over two seconds while holding level — and its endpoints are solved against the loop's MEASURED 53.0 Hz rather than the sample's nominal 49.8, which would have put it a semitone out. Deep drops measure +7.6 dB of 18-35 Hz over an ordinary bar. The constant sub drone moved off ROOT-12 to sit on the tonic while this was done: at 20.6 Hz it had been the loudest band in the whole bed, spending headroom on a frequency almost nothing reproduces and parking a permanent tone exactly where the sweep was trying to land.", "bytes": 1613408, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain); Calf Studio Gear — Motor (engine noise bed) — Calf Studio Gear (Public Domain); Calf Studio Gear — Static (vinyl/amplifier hiss bed) — Calf Studio Gear (Public Domain)", "license": "CC-BY-SA-3.0" }, "ro_schizzle": { "src": "music/ro_schizzle.ogg", "ms": 137143, "loop": true, "bpm": 84, "key": "F1 minor blues", "notes": "Shop rotation. THE SCRATCH METHOD CROSSED WITH PURE INSTRUMENT SAMPLING. Every scratch on this record is synthesised rather than found: there is not one usable scratch sample on this machine (the only 808 and Hydrogen kits present are GPL and the licence gate refuses them), so scratchkit drags a playhead across source audio instead — a scratch is not a timbre to be recorded, it is a function from output time to read position, which is why pitch and duration move together exactly as they do under a needle. The source is the MC's OWN three takes, which is both what a DJ actually does with a vocal stab and free of any provenance question because we generated it: fifteen one-shots, five gestures (baby, chirp, transformer, stab, tear) across three takes, 84 cuts placed. THE PIANO IS A SAMPLED INSTRUMENT PLAYED BADLY ON PURPOSE and rendered as its own stem so it can be treated as a sample rather than as a track — the character of a hardware sampler lives in the sample, because degradation before the arrangement sounds like a record made on the box while the same processing over a finished mix sounds like a plugin. So the stem goes through an SP-404 model (band-limit in, sample-rate hold, 10-bit quantisation referenced to its own peak, reconstruction filter out) and is then wowed at 0.55 Hz, about one cycle every two bars, so the wobble never syncs and never resolves. What makes the PLAYER drunk is in the score and none of it is an effect: the timing jitter is asymmetric (-14 to +42 ms) because a bias toward late reads as dragging while symmetric jitter just reads as bad quantisation; each hand draws its own jitter so the hands disagree with each other and not merely with the grid; velocities are random and unrelated to the beat so no accent pattern emerges; and the occasional wrong note is always a semitone BELOW and always struck early, because that is the miss a real hand makes falling off a black key onto the white one under it. 604 notes. THE ARC IS THE OWNER'S, STATED ALMOST AS A SPEC: one measure of ubertempo death metal, then the dirty scratching and the filthy bass. Bar 1 is a genuine 32nd-note blast — 672 hits a minute at 84 bpm — on the tight kit with a tremolo wall over it, and bar 2 drops the floor out into the groove; the gag returns at bars 17 and 33, so the record keeps threatening to become a death metal album and keeps changing its mind. Measured on the finished file, the insanity measures sit 17.3 dB above the groove in the 2-8 kHz band, which is what a cymbal wash and a tremolo wall actually do to a spectrum. The bass is the pre-domed 808: tanh is an odd function and puts the second harmonic of a 35 Hz sine at the noise floor, which is exactly the octave a phone passes, so an asymmetric shaper makes it once, in the right place, and instr 25 only glides it — 30-70 Hz sits 1.4 dB under full mix level. Three one-liners from an MC, deliberately NOT pitch-corrected because an MC is whatever he is, landing 17-20 dB over the bed. A room of people hooting and hollering underneath, gated from BOTH sides at 8-15 dB under the music, because background has a floor as well as a ceiling and a crowd 25 dB down is not background, it is absent.", "bytes": 3177345, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "eight_drunken_immortals": { "src": "music/eight_drunken_immortals.ogg", "ms": 116044, "loop": true, "bpm": 91, "key": "C#1 aeolian (the sway is not the strike)", "notes": "Boom bap with a systematic stagger, and every melodic note played twice. THE SWAY IS NOT THE STRIKE. Drunken boxing is not a man failing to stand up — the instability IS the technique, the body lurches so the strike arrives from somewhere unwatched, and the strike itself lands exactly where it was aimed. That is a claim about TIMING, so this record makes it instead of referencing it. Every voice carries a per-sixteenth offset drawn from a bounded random walk, the walk is multiplied by sin(pi*s/16) so it is EXACTLY ZERO on the downbeat and closes back to zero as the bar turns over, and the offset is then scaled per voice: the kick and snare are pinned at 0.00 while the hats sway at 1.00 and the hand percussion at 1.15. Rendered in isolation and timed against the swung grid, the kick lands with a standard deviation of 0.22 ms and a worst case of 0.4 ms; the hats scatter at 34.23 ms, mean 21.45. The hats are 157 times as unsteady as the kick, and the downbeat never moves at all. EIGHT IMMORTALS, eight characters of lurch, one per five-bar section, reaches running 0.05 up to 0.23 and back down to 0.06 — the record gets drunker through the middle and sobers up to land the last one. FIVE bars against a FOUR-bar bass figure is the Aesop Rock end of it: the cycles realign only every twenty bars, so an immortal never enters on the same beat of the bass pattern twice running, and the forty-bar body is exactly two super-cycles. SAMPLED AND RENDERED ON THE SAME NOTE. Every melodic line is played twice — once by a recording from the public-domain GM bank, once by a synthesised voice from the orchestra — at one pitch and one staggered instant, with Karplus-Strong under the piano, inharmonic FM under the vibraphone, a filtered pulse under the muted trumpet and a saw bed under the strings. It is enforced by the call site rather than by discipline: `overlay()` is the only way a melodic note gets written, and the two halves cannot drift apart in pitch because the sampled voice takes an integer MIDI note (sfplay3 truncates, so it has no choice) and the rendered voice is handed midi_to_hz OF THAT SAME INTEGER. It is the honest inversion of what RZA was doing: he layered records on records, and with no records to layer this layers the recording on the synthesis. The one voice with no sampled twin is the Cypress Hill whine — a plucked string through the modelled amp, played high and thin, because nothing in the bank sounds that unpleasant. C# AEOLIAN, TONIC 34.65 Hz, inside the 30-45 Hz band a car cabin amplifies as a pressure vessel. THE VOCAL IS ORIGINAL AND THAT IS DELIBERATE: the brief named an ODB reference, a style is not ownable and a lyric is, so the take borrows the delivery and writes its own words — and it reaches the record only through the crossfader, cut into ten scratches alongside a cackle that has now laughed on four tracks. Nothing here samples anybody's record. Measured twice and corrected once: the melodic bus first contributed +0.5 dB overall and left the eight sections 1.6 dB apart — an arrangement that adds vibes, a horn, a guitar, strings and a choir, and is FLAT — which is QUAL-323 exactly, so it was re-solved from a mute test to +6.2 dB at 1.2-4 kHz and a 4.2 dB spread.", "bytes": 1545510, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain); Calf Studio Gear — Static (vinyl/amplifier hiss bed) — Calf Studio Gear (Public Domain)", "license": "CC-BY-SA-3.0" }, "chrome_relic_minigun": { "src": "music/chrome_relic_minigun.ogg", "ms": 118000, "loop": true, "bpm": 80, "key": "Eb1 phrygian dominant (80/96/120 on one pulse)", "notes": "Pantera and Denzel Curry, and a tempo that moves without moving. THE PULSE NEVER MOVES. \"Tempo starting out 80 and going to 120, yet still matching the rest of the song's rhythms\" sounds like two requirements that fight, and it is not — 120/80 is exactly 3/2, so a TRIPLET eighth at 80 BPM and a STRAIGHT eighth at 120 BPM are both 0.25 s. There is a quarter-second pulse, 240 a minute, that exists identically at every tempo on the record; the tempo does not accelerate, the SUBDIVISION changes underneath a pulse that is standing still. That is metric modulation, and it makes \"still matching\" a fact rather than an aspiration: nothing has to be nudged into place because the grids share every fourth of a second. The ladder is 240 divided by 3, 2.5 and 2 — at 80 a bar is 12 pulses, at 96 it is 10 (a HALF is five, which is why the middle section lurches), at 120 it is 8. Every bar at every tempo is a whole number of pulses, asserted at import, and the pulse is not a theory: a rimshot plays it 472 times from the first bar to the last, every gap exactly 0.25 s, so you can hear the thing underneath refusing to change. Measured on the finished master by autocorrelating the onset envelope, the 0.25 s period is the strongest one in every section and beats the bar period in six of seven. THE DRUMS ARE NOT MONO-RHYTHMIC BY CONSTRUCTION: the riffs are written in SIXTEENTHS OF WHATEVER BAR THEY ARE IN, so the same pattern is a different absolute rhythm at 80, 96 and 120 without anyone writing three sets of parts, and six kit patterns rotate underneath — the snare plays a figure while the cymbal carries the speed, because a snare on every subdivision is a wash and not a rhythm. The kick is DERIVED from the riff (a quarter-note anchor unioned with the riff's steps) so a chug can never land where a kick is not. THE GUITAR HOLDS ONE SHAPE, root and fifth, and every other pitch comes from a baked whammy bend — the same rule as its predecessor. The SQUEAL is the orchestra's pinch-harmonic mode, instr 17 with a negative slide parameter, which makes the string JUMP to a harmonic instead of gliding to one; that is what a pinch actually is, and played as a line rather than as punctuation it is the Megadeth half of the brief. The 808 is ruined asymmetric-first then filtered, because the shaper needs a whole waveform to make even harmonics out of. Eb PHRYGIAN DOMINANT — a flat second over a MAJOR third, the mode half the Megadeth catalogue is built on — with the tonic at 38.89 Hz, inside the band a car cabin amplifies. Four rap bars, two bars each, all inside the 80 BPM verse: the placement model has one bar length and a three-tempo record does not, so a line placed with the wrong bar length would not land slightly off, it would drift further off with every bar. ONE BUG WORTH RECORDING: the pulse was scheduled 472 times and rendered as SILENCE on the first pass, because the sample pack emits an f-statement only for tables that have already been requested and the table list was built before the events were — the rimshot's number was allocated afterwards and never declared. Csound plays a note pointing at an undeclared table as nothing at all, with no error. That is QUAL-317 in a new costume, and a render-time assertion now compares every instr-20 event against the declared set. AND THE RAP WAS DROWNING THE MUSIC, which took two measurements to fix rather than one. Aligning the bed and the finished master on 40-120 Hz — a band the voice cannot reach — and subtracting, the vocal sat 9.0 dB above the music in its own band (1.2-4 kHz), where a rap vocal wants roughly 4 to 6. Two separate reductions had compounded: the verse kit was ALSO pulled down to make room, so the music was already 3.9 dB below the bridge before a voice was added. Both were corrected — the kit level came back up, the vocal target came down 3.6 dB — and the residual gap was left alone on purpose, because it is the verse's ARRANGEMENT being sparser (half-time kits, the stomp and halt riffs) and that is what a verse is supposed to be. The voice now sits +5.3 dB over the music at 1.2-4 kHz and 2.5 dB UNDER it at 300 Hz-1.2 kHz, where the guitar and bass live. THE VOCALS WERE RECAST. Every line in the verse now comes from the CIGAR narrator rather than the hype shouter, because a boast shouted by someone visibly enjoying himself is just a boast — the comedy is the gap between the weight of the voice and the size of the claim. Voice-design is a SAMPLE FROM A DISTRIBUTION, not an instruction: one unchanged description asked four times produced four different people, and measured on voiced frames the share of energy below 130 Hz came out 52.6, 0.5, 0.0 and 0.0 percent — three takes with no bottom at all, none of which errored. Naming the REGISTER explicitly in the delivery note (the first version described the performance and mentioned pitch once, in passing) moved the median from 12 to 27 percent, and the takes are then chosen best-of-five rather than against a threshold, because the threshold that was there first had been set from a single lucky draw and five subsequent draws never reached it. The deepest reads are also the SLOWEST — a lower, heavier delivery takes longer to say the same words — so the uniform two-bar grid was throwing away exactly the takes most in character, losing the best of the whole run by 0.55 s. The arrangement moved instead: every cue is now placed from its take's measured length. Two answers close the joke, a beat and a half after the last boast, from two different women so it reads as a room reacting rather than one person repeating herself. THE ANSWER BECAME A ROOM. One woman replying became four, each a different person rather than one voice four times — measured on the takes they run 154, 195, 262 and 314 Hz with speaking rates from 5.3 to 9.6 syllables a second, and no two of them could be the same woman. There are only two female treatments in the cast, so the chains alternate: adjacent cues differ in BOTH voice and processing instead of leaning on the voice alone. The narrator keeps the centre and they answer from around him, which is the staging the joke wants — one man being appraised by a room, not four voices in a queue. He and the droid chuckle underneath afterwards, overlapping by a second, because two people laughing at one joke start together and trail apart; sequencing them would make the machine a reply rather than the pair of them agreeing. The droid's pitch is now derived from a SUSTAINED PROBE at the same espeak settings rather than measured off the laugh itself: a chuckle is three short bursts with no stable fundamental, and measuring one returned 52.6 Hz for a file with no energy anywhere near it. Pitch is a property of the voice settings, not of one utterance.", "bytes": 1614505, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain); Calf Studio Gear — Static (vinyl/amplifier hiss bed) — Calf Studio Gear (Public Domain); Calf Studio Gear — Motor (engine noise bed) — Calf Studio Gear (Public Domain)", "license": "CC-BY-SA-3.0" }, "down_for_robotics": { "src": "music/down_for_robotics.ogg", "ms": 115200, "loop": true, "bpm": 200, "key": "D2 locrian (one fret, one lever)", "notes": "Brutal metal at 200 BPM, and A CALLBACK: \"Put it down for Robotics\" is the four bars where 'Process of Destruction' — three rows up this page — detonates into a metal riff. This is that instant given its own record. THE FRETTING HAND NEVER MOVES. Every guitar sound here is the SAME NOTE, D2 at 73.42 Hz, and every other pitch the guitar produces comes from the whammy bar: no riff, no fingering, one fret and a lever. That is a real technique rather than a stunt — drop-tuned one-note chug riffs hold most of a modern brutal record together and the dive-bomb has been the genre's melodic instrument since Van Halen — and doing BOTH and nothing else is the constraint. It is asserted at render time in two directions: every guitar event carries the same note number, AND every baked bend plays at playback ratio exactly 1.0, because instr 20's p5 is a pitch control and transposing a bend there would be a pitch change that did not come from the bar. Measured on the finished divebomb by spectral centroid — autocorrelation cannot track a distorted dyad, which has no single fundamental — the pitch travels 1748 Hz down to 338, a net -28.4 semitones against -26.0 asked, at a median 0.17 semitones per 23 ms frame. That is a GLIDE. A fretting hand would step and then hold. THE HARMONY THEREFORE LIVES ENTIRELY IN THE BASS, which is why the mode is D LOCRIAN: its diminished fifth is in the scale rather than borrowed, so the 808 walks through the tritone against a guitar frozen on the root and the dissonance is structural instead of decorative. Locrian is barely usable and that is the point. THE 808 IS RUINED ON PURPOSE — asymmetric saturation first, then a filtered dirt stage, in that order because the shaper needs the whole waveform to make even harmonics and filtering ahead of it would remove exactly what it works on. Normalised to equal peak, the treated kick gains 13.8 dB at 60-120 Hz and 30.7 dB at 120-300 while 30-50 Hz moves only 2.0: nasty upstairs, sub untouched. The kit blasts at sixteenths, 13.3 hits a second, on the tight kit because a room kit at this tempo is a wash and not a kit; the breakdowns drop to half time on the roomy one, because the whole point of a breakdown is that it is SLOWER than the thing it interrupts. THE VOICE is CIGAR, new to the cast: thirty a day for thirty years, narrating like a film trailer with a growl underneath every word. Two jobs that pull against each other — narration is enormous and clean, a growl is dirty — so the size is built with EQ and the dirt with a soft-clipper rather than by driving the compressor, which would flatten the slow swell that makes narration sound narrated. The model would not render below about 110 Hz however the prompt was written (rewriting the growl's direction to name the physical mechanism moved it 2.8 Hz, which is nothing), so the roar is pitched down to 82 Hz afterwards with the formants left to fall with it — the speaker gets physically bigger, which is how trailer voices and death-metal vocals are actually made. Three words, three times, each on the downbeat of a half-time bar, because that is the only instant with room in it: measured on the finished master the lyric bars carry 8.3 dB more energy in the 300-3000 Hz voice band than a blast bar does. The grunts sit off the downbeat on purpose — a \"Yeah\" landing with the kick reads as part of the drum pattern rather than as somebody in the room. RE-CUT after the first pass, on three counts. THE RIFF AND THE KICK ARE NOW ONE OBJECT. A record with a single chord shape has exactly one axis left to be interesting on, so there are nine riffs turning over every four bars — and `riff_of(bar)` is read by the guitar AND by the kit, with the kick's step set being the blast's own even sixteenths UNION the riff. A chug can no longer land where a kick is not. The first cut kept those as two hand-maintained lists and two hand-maintained lists drift: only 62% of chugs had a kick underneath, and the other 38% were landing on a snare. Rendered in isolation and timed against the scheduled kicks it is now 100% inside 3 ms, median 0.92. The direction of the union matters — forcing the riff onto the kick would make every riff straight eighths and the record would have one rhythm, whereas letting the KICK take the riff's odd steps keeps the blast running and still guarantees the lock, which is also what a drummer actually does with a guitarist. THE CHUGS ARE POWER CHORDS NOW, root and fifth, and the rule survives because a power chord is ONE SHAPE on two strings rather than two notes — the hand still never changes position. AND THE MIX WAS HARSH FOR A REASON WORTH NAMING: isolated over one section, the snare was the loudest source at BOTH 2.5-6 kHz and 6-12 kHz, because 92 full-velocity backbeat cracks in sixteen bars at 13.3 a second stop being hits and become a wash. Velocity is timbre on an AVL kit — five recorded layers — so the blast fill is now played light at vel 0.48, which is a genuinely darker RECORDING and not a quieter copy, and a real backbeat went back on 2 and 4 to carry the definition the wash was pretending to provide. There had been no backbeat at all: steps 4 and 12 are even, so the kick always took them. Against the 41-track catalogue the record went from +3.3 dB at 6-12 kHz to +1.7 and from +1.1 at 2.5-6 kHz to -1.2. The guitars were then panned hard, which is where a metal rhythm track belongs, after checking that two near-identical parts 7 ms apart still sum to mono without cancelling: they do, at -0.0 dB, because 7 ms puts the first comb null near 71 Hz, below the guitar's range. RE-CUT AGAIN after the owner named the two passages that worked — 37-42 s and 1:00-1:11 — and asked for a beat that was not 'bam bam bam bam'. Both complaints had the same cause. ONE beat ran under all 48 blast bars: kick on every even sixteenth, snare on every odd one, identical every time, which at 13.3 hits a second is not a rhythm but a texture. There are now seven kit patterns plus the half-time one, rotating every four bars, and the sixteenth-note pulse moved OFF the snare and onto the cymbal — which is how fast metal actually grooves, the snare playing a figure while a ride carries the speed. The kick's anchor also changed from every even sixteenth to QUARTERS, so the riff library finally does the work it looked like it was doing: with the old anchor the kick pattern was identical in every bar no matter which riff played over it. Eight beats across twenty distinct beat-and-riff combinations, the busiest holding a quarter of the record instead of one holding all of it, and the true blast beat cut to four bars so it lands when it arrives. THE TWO PASSAGES TURNED OUT NOT TO BE ABOUT VARIETY AT ALL. Measured, those bars are MORE similar to each other bar-to-bar (0.960 rhythmic similarity) than the body average (0.925); what they share is a half-time breakdown on the roomy kit with a whammy bend over it — space, not novelty. So every eighth bar of a blast section now drops to half time with a bend across it. Roomy bars went from 20% of the body to 28%, bars carrying a bend from 12% to 20%, and the longest stretch with no breath at all is 10.8 s. On the finished master those breath bars are the LOUDEST bars on the record, which is the point. Vocal clearance improved to +10.2 dB and the crest sits at 12.2. One measurement lesson is recorded in the source: the presence bands are owned by the BENDS, two of which outweigh 416 chugs, because a baked bend is a 1.6-bar sustained chord and a chug is a 62 ms stab — two rounds of trimming snares and cymbals moved those bands 0.1 dB each before anything was isolated and actually measured.", "bytes": 1523614, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain); Calf Studio Gear — Static (vinyl/amplifier hiss bed) — Calf Studio Gear (Public Domain); Calf Studio Gear — Motor (engine noise bed) — Calf Studio Gear (Public Domain)", "license": "CC-BY-SA-3.0" }, "scrap_baron": { "src": "music/scrap_baron.ogg", "ms": 87273, "loop": true, "bpm": 132, "key": "F#1 phrygian (a drone that is a struck string)", "notes": "A boss theme whose DRONE IS A GUITAR NOTE THAT WILL NOT DIE. Every drone in this collection so far has been a pad, a sub or a held sample — something that sustains because sustaining is what it does. This one is a STRUCK STRING that refuses to stop: the amp model was given feedback, expressed as negative damping on a natural-log parameter, so a note can be told to last almost indefinitely without the model running away. Eight notes carry eighty-six of this record's eighty-seven seconds, against four hundred and thirty-two riff notes averaging a tenth of a second with no feedback at all — asserted in both directions, because a record where everything sustains is a wash. It matters musically rather than technically: a pad has no attack and no player, but a bloomed string was PICKED, so the drone has a gesture at the front of it and a hand behind it, and because the sustain lives in the string rather than in a sample THE WHAMMY STILL WORKS ON IT — the drone bends while it rings, which is the one thing a held sample can never do. The drone is a POWER CHORD, root fifth and octave through one saturator so the distortion intermodulates them into a wall, and it runs through the auto-wah, which the guitar could not reach until this track needed it to: a bloomed power chord swept by a wah is the psychedelic half, the whammy over the top is the thrash half. Trap hats sit over a half-time boss kit — fast subdivision against a slow pulse, which is the trick metric modulation plays, done with ornament instead of arithmetic. The bass is the harmonic-series descent borrowed from LOOSE ON THE PLASMA, 4f 3f 2f f stepped down and glided to the floor, at four of the turns. One line from the narrator, and its level was not chosen by ear: a new measurement tool read it at 10.7 dB over the music in the band a voice competes in, asked for a 7.7 dB trim, and the corrected master landed at 3.0 — one pass, where the previous record needed six.", "bytes": 1139880, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Blonde Bop” by Glen MacArthur, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "junk_gang": { "src": "music/junk_gang.ogg", "ms": 123429, "loop": true, "bpm": 140, "key": "D1 phrygian (a 3-and-5 walk, and the sway is not the strike)", "notes": "PLAYLIST STUDY 7, FORKED — the same rhythm section, and the story it did not have. The owner's verdict on `scrap_gang` was that the groove and the dirtied drums were right and the record was \"almost there — it just needs a musical story woven in\". So nothing that was working was touched: `masters_junkgang/{dirt808,scratches,metal}` are BYTE-FOR-BYTE COPIES of that record's prep artifacts rather than a re-run, so the 808's saturation, the 41-slice chopped metal breakbeat and the guitar-sourced scratches are the same recordings and not a second derivation that might land a hair different. The riff table, the kick/riff union, the one-displacement kick-chug fusion and the transient shaping are ported verbatim; the lock still measures 100% inside 3 ms at a median of 0.02 ms. WHAT WAS WRONG WAS AN ARRANGEMENT HOLE, AND IT WAS FINDABLE BY READING RATHER THAN BY LISTENING. `scrap_gang` generates ten scored variations of its four-note cell and then plays none of them. `MO.variations(HOOK, 10)` is called, `VARS[0]` is used once in bar 3, and the function that builds bars 9-68 — every bar of the body — accepts a `vary_from=` parameter that its own body never reads. The cell sounds in bars 1-8 and 69-72 and NOWHERE ELSE: sixty of seventy-two bars are a rhythm bed with a drone over them. The ideas were generated and discarded, which is why the record reads as unfinished rather than as thin. Measured on the shipped masters, that shows up as a body whose melody band (300-1200 Hz) spans just 2.0 dB across seven sections — it does not develop, it alternates. This one spans 4.7 dB. THREE VOICES FILL THE HOLE, EACH ANSWERING A NAMED PART OF THE BRIEF. THE WHAMMY IS A CHARACTER, NOT AN EFFECT. `scrap_gang` fires the bar once, a dive bomb at bar 55, and leaves four of its five shapes unused. Here there are ten placements and the depth climbs 2 to 14 semitones across the record, so the same gesture reads as one hand getting more insistent: scoop asks, warble doubts, rise lifts, dive threatens, flutter argues. THE STRINGS ARE DRUNK, AND THAT IS A TIMING CLAIM RATHER THAN A TIMBRE ONE. The Wu-Tang half is the sampled GM section played as short dyad STABS — never as a pad — doubled by a synthesised drone underneath at 0.30, because the public-domain bank's sustains are ten-frame loops and a held section note becomes an audible oscillator without a body under it. The drunk half is the stagger from `eight_drunken_immortals`: a bounded random walk per sixteenth multiplied by sin(pi*s/16), so it is EXACTLY ZERO on the downbeat and closes back to zero as the bar turns. THE STRIKES LAND AND THE CREW LURCHES — kick and snare are pinned at lurch 0.00 while the strings run 1.05, worst mid-bar displacement 72.4 ms against 0.0 ms on every downbeat. That is drunken-master timing, where the instability is the technique and the strike still arrives where it was aimed; it is not sloppiness, and the difference is asserted rather than described. THE BASS WALKS, AND THAT IS THE CHANGE A CAR HEARS. `scrap_gang`'s 808 plays MIDI 26 and nothing else, with a 29 every fourth bar — the flattest element on the record, and the reason it sits rather than travels. Here it climbs two ladders whose every step is exactly a minor third or a perfect fourth and whose every rung is a degree of D phrygian: D1-F1-Bb1-G1 cruising, and D1-G1-Bb1-Eb2-C2-G1 lifting. Both close back on the tonic, so the figure rolls instead of stopping, and both are asserted at build time in intervals and in mode, because a promise nothing measures is exactly what `vary_from=` already was. Two rungs a bar — the downbeat and the and-of-3 — makes it a half-bar lope rather than a sixteenth pattern, which is the difference between a bassline that travels and one that busies; the low ladder turns over every two bars and the high one every three, and neither divides the four-bar riff cycle, so the loop never predicts itself. The measurable consequence is the point: against `scrap_gang` at identical length, peak and loudness, the sub band (20-60 Hz) falls 0.6 dB while 60-120 Hz rises 1.2 — the walk moved energy off a static 36.7 Hz tonic and into the band a car cabin and a phone can actually reproduce. Mid rises 2.2 dB and presence 1.6, which is the story arriving where melody lives. THE FORM DEVELOPS INSTEAD OF ALTERNATING. The parent is two states taking turns, A B A B A, and nothing arrives that did not arrive before. This runs ten sections and EVERY SECTION CHANGES EXACTLY ONE THING, which is what makes an arc legible on a loop a player is not concentrating on — the ear notices one new voice and does not notice three. The yard states the cell alone and the whammy asks its first question; the bass starts moving; the crew arrives drunk; the teeth scatter them, and the walk keeps running underneath so the heavy bars are part of the journey rather than a pause in it; the ride lifts to the high ladder; strings and guitar trade two bars each through the argument, which is the melodic peak at +6.1 dB; the second wall strips everything back for the dive bomb; the bloom puts all three voices together; the room empties for four bars — a real 1.3 dB drop, where the parent's comedown was as loud as everything around it — and the cell returns exactly as it began. 123.4 s, 72 bars, D phrygian, loop seam 0.87 against a house gate of 12.0, eight loudness passes, no clipped samples. NOTHING HERE WAS HEARD. Every claim above is a measurement — band shares, timing histograms, per-section energy — and whether the whammy reads as one voice arguing, whether the strings sound drunk rather than late, and whether the bass actually bounces are judgements no instrument in this pipeline can make.", "bytes": 2975803, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "scrap_baron2": { "src": "music/scrap_baron2.ogg", "ms": 87273, "loop": true, "bpm": 132, "key": "F#1 phrygian (a five-beat radar ping over a struck-string drone)", "notes": "SCRAP BARON, AFTER A STUDY OF XAVIER WULF — and only two things from that study are in this record, because a study that changes everything is not a study. Wulf is Memphis, ex-Raider Klan and Seshollowaterboyz, one of the people trap-metal is usually traced back to; the 90s Memphis inheritance is hi-hats and 808s over a dark room, which `scrap_baron` already had, and repeating it would have been a coat of paint. The two findings that were NOT already here are the ones that landed. THE SUBMARINE-RADAR PING. The most-cited texture in his production is a sonar-like beep: one high tone, absolutely regular, sitting over everything and never developing. It works because it is the only element in the mix that does not RESPOND to anything — the drums move, the drone bends, the riff turns over, and the ping is simply always there, like an instrument reading a number off a dial. On a boss theme that is a gift, because it is what a targeting system sounds like. THE PERIOD IS FIVE BEATS, WHICH IS THE WHOLE TRICK: five does not divide a four-beat bar, so the ping lands on a different beat every bar and only returns to the downbeat every fourth. Regular and never aligned — the ear cannot file it as part of the beat and cannot dismiss it as random either. A ping on beat one would be a metronome and would vanish inside a minute. It is instr 11, the inharmonic bell, because a radar return is not a musical pitch and a sine would read as one; each ping gets a quieter answer a beat and a half later, panned opposite, since a single ping is a beep and a ping with a return is a MEASUREMENT of something at a distance. RAPID 808s AND SLOW BASS, INTRODUCED SEPARATELY. Not one bassline with a fill: two instruments with different jobs arriving at different points in the record. The slow bass is the harmonic floor and is present from bar one. The rapid sixteenth-note 808, syncopated off the kick, DOES NOT EXIST until bar 17 — and when it arrives it is an event rather than a texture, which only works because the ear has had sixteen bars to learn the record without it. `scrap_baron` had one bass doing both jobs adequately, and splitting them is the biggest structural change here. THE VOICES ARE BREATH, NOT WORDS. `scrap_baron` carries exactly one spoken line and the arrangement is built around the hole it sits in, so a lyric would compete with it. This adds NON-LINGUAL vocalisations only — CIGAR humming, grunting and growling from the chest, and two women sighing, humming and breathing close to the microphone. It is also the honest use of a TTS: a model asked for a word makes a word, but asked for a grunt it makes a SHAPE, and a shape survives being buried where a syllable turns to mud. CIGAR anchors the section walls because he is the weight in the room; the women sit inside phrases and off the beat, because a breath on a downbeat is an event and a breath between them is a presence. THE BRIEF WAS \\u201cdon\\u2019t overdo it\\u201d, AND THE RENDER TREATS THAT AS A NUMBER. Fourteen placements across forty-eight bars, and the mixer asserts the layer sits between 15 and 27 dB under the bed. It caught the first attempt doing exactly what the brief forbade: at a gain of 0.34 the vocalisations measured -6.6 dB against the music, which is a DUET — two voices in the room, the breaths competing with the narrator this record is built around. At 0.075 they measure -19.7 dB: heard, not attended to. The gate was not widened to admit the loud version. Everything else is `scrap_baron` unchanged: the struck-string drone with feedback that will not die, the whammy bending it while it rings, the power chord through the auto-wah, trap hats over a half-time boss kit, the harmonic-series descent. Master trim 0.72 to 0.64 because the ping and the rapid 808 took the peak to 1.094 with 17 clipped samples and the Csound master soft-clips SILENTLY; ceiling -7.5 rather than the house -5.0 because a wall of drone publishes quiet. Loop seam 1.24 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2082263, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Blonde Bop” by Glen MacArthur, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "nasty_on_delayed_arrival": { "src": "music/nasty_on_delayed_arrival.ogg", "ms": 102171, "loop": true, "bpm": 150, "key": "D1 (jungle two-step, amen and roller across a rising tempo)", "notes": "STANDARD JUNGLE DRUMS, AFTER A CLEVERER IDEA FAILED IN THE ONLY TEST THAT COUNTS. `nasty_on_arrival` climbs 150 to 175 to 200 and back, section by section, and that tempo plan is good; what happened INSIDE a section was one 32-step pattern looped for the whole section, so the tempo was the only thing that ever developed. The first attempt at fixing that built a EUCLIDEAN LOOM: four layers of E(k,n) with pairwise-coprime onset counts, each rotating on its own coprime schedule, provably never repeating a bar inside a section. It was mathematically sound and the owner's verdict on hearing it was that it was a total regression. Both of those are true at once, and the reason is worth keeping rather than quietly deleting: four maximally-even layers that agree on almost no step produce a texture with no PULSE in it. Every onset is equally unexpected, so nothing is a downbeat, and a drum part whose job is to be the floor cannot be the least predictable thing in the mix. Interference is a good way to generate detail and a bad way to generate groove. WHAT REPLACED IT IS DELIBERATELY ORDINARY: the two-step, the Amen skeleton, a roller and a half-time — the patterns jungle and drum and bass actually run on. They are decades old and they work because the kick and the snare state the metre plainly while everything interesting happens in the ghosts around them. VARIATION COMES FROM THE GHOSTS, NOT THE SKELETON: the kick and snare are identical in every bar of a section, which is what a pulse IS, while the ghost snares, the open hat and the ride move on a four-bar cycle and a tuned tom turns the phrase around every fourth bar. That is how a real break varies — the drummer keeps time and decorates rather than reinventing the bar. The ride carries the sixteenths above 175 bpm, because that is where drum and bass's speed actually lives: on the cymbal, while the snare plays a figure. The record already accelerated into DnB territory, so the tempo plan needed no change at all — it needed drums that know where beat one is. The 808 still WALKS a phrase with a per-note sustain rather than hammering the tonic, because the bass was never the problem: a low end with one pitch and one duration is a kick drum with a tail. Two numbers moved that were not compositional, both measured rather than guessed. The master trim sits at 0.54 because 0.58 peaked at 1.017 with five clipped samples and the Csound master soft-clips SILENTLY — it would have shipped as a subtly crunchy record with nothing in any log to say why. And the mastering ceiling is -7.5 rather than the house -5.0, because jungle drums are transient-dense and published loudness is set by the master's loudness-to-peak distance: at the default this published -13.5 LUFS against a -12.6 target, a decibel under its neighbours, which a listener hears as broken rather than as quiet. The gate named the ceiling and the gate was not widened. Loop seam 0.198 against a house gate of 12.0. NOTHING HERE WAS HEARD — which is exactly why the loom shipped once before the owner could say otherwise.", "bytes": 1432765, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Blonde Bop” by Glen MacArthur, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "dont_touch_the_trim": { "src": "music/dont_touch_the_trim.ogg", "ms": 109091, "loop": true, "bpm": 132, "key": "C#1 aeolian (7/8-7/8-4/4-5/4 on one pulse; two subdivision ramps)", "notes": "`rig_etiquette` GIVEN AN ARC. Same song — same shifting meters, same sludge riff, same two voices refusing to agree about truck parking — with one gesture performed twice, at a third of the way through and at two thirds, which is where a listener's attention needs something to happen. The render asserts both land there: 0.31 and 0.65 of the record. THE GUITAR HAS FOUR VOICES AND THREE OF THEM ARE FORESHADOWING. A record that saves everything for its climax has a boring first minute. CHUG is palm-muted power chords, short and dry, under almost everything — the floor, and a palm mute is really a percussion instrument that happens to have a pitch. TWANG is the clean guitar in the quiet bars: what the record sounds like when nothing is wrong, which is what makes the rest mean anything. SCREECH is a real pinch harmonic — instr 17 reads a NEGATIVE RATE as \\u201cdrive the string at a higher partial\\u201d, physically what a pinch harmonic is, a finger at a node so the string vibrates in n segments instead of one — sprinkled early so the ear has met it before it matters. FURY is the wall, and it is TWO BARS of the whole record: measured, 2.1% of the guitar events. It only lands because everything around it is muted, and that ratio is asserted rather than eyeballed. THE RAMP IS SUBDIVISION, NOT TEMPO. Four bars before each detonation the kit climbs eighths, twelfths, sixteenths, twenty-fourths WHILE THE CLOCK NEVER MOVES. The ear reads acceleration because events arrive more often; nothing has sped up. A real tempo change would announce itself and spend the surprise early. The score expands with it — two notes, then three, four, five, climbing the scale — so the harmony opens as the rhythm tightens. That the expansion is MONOTONIC is asserted: the first cut had two ramp bars with the same note count, which is a bar that does not escalate, and the ramp is the entire gesture. AFTER EACH WALL, THE CALLBACK. \\u201cDown for robotics!\\u201d — the same recording from that track — arrives echoed and SELF-DUETED, over steady palm-muted chugging, the harmonic-series deep drop borrowed from LOOSE ON THE PLASMA (4f, 3f, 2f, f: a descent that is harmonic in the strict sense rather than merely downward), scratching on the dotted eighth, and the wasteland women laughing a long way back. THE SELF-DUET IS A NINETY-DEGREE PHASE SHIFT AND THAT IS NOT A DELAY, which is the one technical claim on this record worth stating carefully. A copy delayed by a few milliseconds is the same waveform arriving twice: summed, it combs, and you get a hollow notched version of one voice. A copy shifted ninety degrees in PHASE — the Hilbert transform, the signal's quadrature — has the SAME MAGNITUDE SPECTRUM and a completely different waveform, so it sums to something twice as wide with no notches at all. Every frequency is decorrelated by the same quarter cycle rather than by a fixed time, which is why it reads as two people rather than as an effect on one. Measured on the finished pair: correlation 0.000 where a delay would be high, and magnitude spectra differing by 0.006 dB where a filter would differ by decibels. Both are asserted, because a phase shift that silently degraded into a delay would sound almost right and be wrong for a reason nobody could hear their way to. The echo is locked to the dotted eighth at 341 ms so the repeats land inside the meter — on a record in 7/8 and 5/4 a free-running delay fights the bar on every repeat. Loop seam 0.009 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2800752, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "chromebeard": { "src": "music/chromebeard.ogg", "ms": 117639, "loop": true, "bpm": 150, "key": "C2 aeolian (150->75->150 bpm, a REAL tempo map; octave descent at the crush)", "notes": "STUDIED FROM GNOME (Antwerp) — “Ambrosius”, from *King*, 2022. Belgian stoner/prog/doom. What the study actually yielded was mostly a list of things NOT to do. IT IS A TRIO: guitar, bass, drums. No keys, no strings, no samples, no 808. Every other record in this catalogue reaches for an orchestral colour when a section needs to grow; this one is forbidden to, so growth has to come from three instruments playing DIFFERENTLY. That restriction is the single most important thing taken from the source, and the render refuses to build a fourth instrument family into the bed — measured, 3 families, asserted. THE HOOKS ARE SILLY AND THE RIFFS ARE NOT: the band's whole trick is mock-grandiose nonsense delivered dead straight by a rhythm section that is not in on the joke. The joke is in the words, never in the playing. FUZZ IS A SUSTAIN DEVICE, WHICH IS WHY THE HOLES ARE THE GENRE. Distortion compresses, so a fuzzed string decays far more slowly than a clean one and a stoner riff can leave enormous gaps and stay loud. Filling those in turns a stoner record into a bad thrash record, so every riff here is gated at 55% of its sixteenths or fewer. THE TEMPO MAP IS THE STRUCTURE, and it is the first real one in this catalogue. Everything else here was written on a pulse that never moved — DON'T TOUCH THE TRIM went out of its way to FAKE acceleration with subdivision ramps precisely so the clock could stay fixed. This record does the opposite and slows down for real: 150 bpm, a two-bar ritardando to 75, eight bars of crush, a two-bar accelerando back. 75 is EXACTLY half of 150 on purpose — the doom section's quarter note is the drive section's half note, so the pulse survives the drop and the ear keeps counting while the ground falls away underneath it. A non-integer ratio would sever the pulse and read as an edit between two songs. RITARDANDO IS GEOMETRIC IN THE PERIOD, NOT LINEAR IN THE BPM. Tempo is perceived as a ratio, the same way pitch is — 60→70 bpm and 120→140 bpm are the same musical event. So the ramp interpolates the BEAT DURATION geometrically, period[i] = P0*(P1/P0)**(i/N), which is a straight line in the log domain, lands exactly on the target, and slows evenly to the ear: 150 138 126 116 106 97 89 82, then 75. A ramp linear in BPM spends too long near the fast end and then lurches, which is the sound of a DAW tempo track rather than of a drummer. That the ramp is geometric is asserted on the ratios between consecutive beats, not eyeballed. THE OCTAVE DESCENT is the figure the whole record is built toward: C3 down through C aeolian to C2, seven scale steps and the octave, touching every degree on the way and landing on the root. Held nearly the full half note with feedback bloom so each chord is still ringing when the next arrives — a continuous fall rather than eight separate events. Played twice: once plain, once elaborated with a scooped bar and a pinch harmonic squealing off the top of each chord. Asserted monotonic, asserted exactly twelve semitones, asserted to land on the root. THE FOUR GESTURES THE BRIEF NAMED ARE ALL ONE INSTRUMENT. instr 26's string is a k-rate delay line, so the score can bend, mute, swell and feed back a note WHILE IT RINGS. Distortion: power() sums root+fifth+octave BEFORE the gain stage, so the amp generates the intermodulation that IS a power chord rather than three clean notes added up. Bending: 65 bent notes across 5 bar shapes — dive, scoop, warble, flutter, rise. Squealing: 40 pinch harmonics, the string driven at its 4th partial, which is physically what a finger resting on a node does. Feedback: 36 bloomed notes, where loop gain becomes 1-(1-kFb)*exp(-bloom) — a natural-log control that approaches unity and can never reach it, so no value diverges and each +1 multiplies the decay by about e. That is why the intro can be one chord for six seconds and still be interesting. THE BAND STEPS BACK FOR THE VOICE, AND THAT IS WRITTEN INTO THE SCORE RATHER THAN THE MIX. The one spoken line lands on the crush downbeat at 62.0 s, 53% into the record, at the bottom of the tempo drop — the emptiest the arrangement ever gets. The first cut CLAIMED that room in its docstring and measured 1.4 dB the wrong way: the mix gate refused to ship a line quieter than the band under it, and the fix belonged in the arrangement, not in a vocal fader. That bar now carries one chord instead of two, one guitar instead of a double-tracked pair, a crash already decaying, and no ride. Measured after: the line sits +5.5 dB over the bed beneath it. THE ECHO REPEATS ON THE BAR LINE. A doom bar is 3.20 s and the take is 1.63 s, so repeats delayed by exactly one and two bars arrive on downbeats with the words long finished — a canyon rather than a smear, enormous without costing the one thing the line exists to do. FIVE VOCAL STATEMENTS IN 118 SECONDS, because the source is sparse and the restraint is what makes the one line land. Loop seam 0.922 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2791251, "credit": "Drums: AVL Drumkits “Red Zeppelin 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)", "license": "CC-BY-SA-3.0" }, "super_haul_haul": { "src": "music/super_haul_haul.ogg", "ms": 116364, "loop": true, "bpm": 99, "key": "G#1 minor pentatonic (upright bass leads; drums LOOP on purpose)", "notes": "STUDIED FROM SOUL COUGHING — “Super Bon Bon”, from *Irresistible Bliss*, 1996. 99 bpm, G#/Ab. The band called their own sound “deep slacker jazz”: improvisational grooves, oddball samples, hip-hop and spoken word, with Mike Doughty's beat poetry over the top. THE INSTRUMENTATION IS THE STUDY, and three of these four are the opposite of what a rock record reaches for. Sebastian Steinberg played an ACOUSTIC UPRIGHT BASS, not an electric one, and it is the LEAD instrument here — the riff is the song and the voice is a guest on top of it; the render asserts the bass plays at least four notes for every sampler stab and sits well above the guitar in level. Doughty's guitar is CHOPPED — muted, percussive strumming, a hi-hat that happens to know the key, so it is written as 60 ms notes at a fifth of the bass's amplitude, and if you can pick out what chord it plays it is too loud. Mark Degli Antoni threw ORCHESTRAL STABS over everything, so the sampler here is the actual 1990s orchestra hit, capped at twenty in the whole record because it works as punctuation and becomes a joke about 1996 if it does not. The kit is a BEBOP kit, because Yuval Gabay is a jazz drummer. THE DRUMS REPEAT, AND THIS RECORD INVERTS A RULE THE REST OF THE CATALOGUE ENFORCES. DON'T TOUCH THE TRIM asserts that no two consecutive bars share their drums, because in that idiom a repeated bar is a composer who stopped writing. Here the percussion has to feel SAMPLED — a jazz drummer imitating a loop — so the bar is written once and repeated, and the gate asserts the REPETITION rather than the variation: 39 looped bars, longest identical run 8. Through-composed drums would destroy the one thing the idiom is built on. All the movement lives in the bass and the stabs instead. SPACE IS THE ARRANGEMENT. Nothing here is distorted and nothing sustains, so two parts playing at once are simply two things at once — density is not thickness, it is clutter. At 10.2 events per second this is the sparsest bed in the catalogue, and that is asserted rather than hoped for. Bars 29-36 carry no voice at all: the bass has led the record since bar one and the break is where that becomes audible. THE MELODY IS DERIVED FROM THE TAKE RATHER THAN WRITTEN, WHICH IS THE REAL WORK HERE. Four records in this catalogue were written the other way round — melody first, voice second — and all four came out demonic or strained at six to twelve semitones of correction. Each time the fix was the same and each time it was applied afterwards, until RIG ETIQUETTE finally added a gate that REFUSES a melody more than four semitones from its singer. A gate is the right safety net and the wrong primary mechanism, because it still lets you write the wrong thing first. So here the home note is not written at all: each take's median f0 is measured, snapped to the nearest tone of G# minor pentatonic, and the line is expressed as offsets in scale DEGREES from its own home. Measured result — hook +0.1, v1 +0.5, v2 -0.7, v3 -0.7, v4 -0.8 semitones. Every line sits within eight tenths of a semitone of where that voice already was, and the register error is not caught but unrepresentable. The gate is kept anyway, because “structurally impossible” is a claim and claims get measured. THAT ALSO SUITS THE IDIOM EXACTLY. Doughty's delivery is near-monotone, so a line has almost no room to move and WHERE IT SITS is nearly the only musical decision there is. Deriving it from the voice is what makes a flat delivery sound relaxed instead of pinched. Every melody here spans five semitones or fewer, asserted. BENDING, NOT RESYNTHESIS. This project has two vocal instruments and choosing wrongly has cost whole rounds: vocalbend resynthesises from an LPC fit of the tract (right for sustained, held, sung material) while autotune.retune_continuous bends the ACTUAL RECORDING (right for material whose character is in the delivery). On shouted takes, resynthesis returned the empty string from ASR on every line while bending the same recordings scored 0.095 WER. Beat poetry is the most delivery-carried material there is; resynthesising it would throw away the only thing that makes it a person. TAU IS 140 ms, DELIBERATELY SLOW. The retune trajectory is an ODE — p[n] = a*p[n-1] + (1-a)*t with a = exp(-dt/tau) — which reaches its target in roughly three time constants. At 40 ms you get the hard modern autotune snap; at 140 ms the pitch arrives a fraction behind the syllable, which is what a talker does and a singer does not. The idiom is TALKING IN TIME, so the correction has to be audible as tuning without ever sounding decided. RE-ROLL BEFORE YOU RE-WRITE — a new finding, filed to the shelf. One take came back at 292 Hz against a 96-155 Hz group: animated where the whole point is flat. Re-writing its delivery note moved it to 213 Hz and stalled, exactly as the existing rule about TTS prose predicts. Sampling five SEEDS at fourteen seconds each found one at 108.4 Hz — eleven semitones, no processing and no argument with the model. The seed-to-seed spread of this model is far wider than its sensitivity to delivery prose; the winning seed is pinned in the render script. THE LYRICS ARE ORIGINAL AND SPARSE — fourteen statements in 116 seconds, in the idiom rather than of the song: the subject/object inversion Doughty favours, CB-radio sign-offs, and a reduplicated nonsense title in the same three-syllable meter as the original's. Tuning measured on the output at 13-25 cents median error per line. Loop seam 0.034 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2833935, "credit": "Drums: AVL Drumkits “Blonde Bop” by Glen MacArthur, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)", "license": "CC-BY-SA-3.0" }, "boost_gauge": { "src": "music/boost_gauge.ogg", "ms": 123429, "loop": true, "bpm": 140, "key": "E1 aeolian (140 written / 70 felt; the wobble RATE LADDER is the riff)", "notes": "STUDIED FROM RUSKO — “Woo Boost” (Mad Decent, 2010), one of the records that defined dubstep's mainstream moment. Described at the time as squeaky, bassy, grinding-into-the-ground and wobbly, with treble arpeggios over the buzz of the bass. 140 BPM WRITTEN, 70 BPM FELT, and that is the grammar rather than a detail. The grid runs at 140 and the drums play HALF TIME — kick on beat 1, snare on beat THREE, never on 2 and 4. At this tempo that leaves 1.71 seconds of air between the kick and the snare, and the space IS the genre. Fill it and the record becomes drum and bass without a single other parameter changing, which is why the build asserts it: measured, 70 snares on beat 3 against 2 on 2/4. THE WOBBLE IS A FILTER, NOT A PITCH, AND THIS RECORD NEEDED BOTH. The owner asked for the synthesiser to be replaced by this catalogue's note-bending guitar. A wobble bass is an LFO on a filter CUTOFF; guitarkit's bar shapes are an LFO on PITCH, and doing only the pitch half produces vibrato rather than wobble. So it is built in two locked layers: the `warble` bar shape supplies the pitch movement and instr 91 — a flanger plus auto-wah, scheduled explicitly in two-bar spans — supplies the actual filter sweep, at the same rate. instr 91 is an INSERT and clears its bus only inside itself, so a voice sending to an unscheduled span accumulates into a global nobody drains; the build asserts every sending note falls inside a scheduled span. 16 spans, 64 sending notes. THE RATE LADDER IS THE RIFF. This is the part that is easy to miss and impossible to fake: in this idiom what a listener remembers is not the notes — the bass sits on E almost throughout — it is the RHYTHM OF THE RATE CHANGES. A drop states a wobble at a quarter note, then eighths, then sixteenths, then triplets, and that ladder is the hook. Written here as an explicit table in beats-per-cycle and converted to Hz so the wobble is always locked to the grid: 1.0, 0.5, 0.333, 0.25, 0.1667 beats per cycle = 2.33, 4.67, 7.01, 9.33, 14.0 Hz. The build refuses fewer than four distinct rates, and refuses a drop with no sixteenth-or-faster wobble in it, because one rate held throughout is a drone with drums on it. THE BASS IS TWO INSTRUMENTS, BOTH IN-HOUSE. A SUB you feel rather than hear, through this project's `domebass` asymmetric saturator at asym 0.72 with a sub-lift shelf — and the asymmetry is load-bearing, because at asym 0 the stage is plain tanh and generates NO even harmonics, and the even harmonics are exactly what let a 41 Hz note be audible on a speaker that cannot reproduce 41 Hz. Then a DIRTY mid-bass at asym 0.45 that keeps its top end and carries the wobble. Splitting them is how a low end stays enormous without turning to mud: one owns 30-60 Hz, the other 80-500, and neither is asked to do both jobs. NEITHER VOCAL IS TUNED, AND THE RESTRAINT IS THE POINT. Every other vocal in this catalogue is bent onto a melody derived from its own measured f0. A drop sample is not a sung note: “Woo!” is 0.56 seconds of shout with no sustained pitch in it, and a retuner works by tracking f0 and moving it. Run on this it would find the tracker's noise floor and bend THAT, turning an impact into a warble. The correct treatment is the one the genre actually uses — place it on the grid, give it a tempo-locked echo so it smears into the drop, and get out of the way. Asserted by absence: the mixer imports no autotune. PLACEMENT IS AN ARRANGEMENT DECISION, NOT A MIX ONE. The shout lands on the last half-beat BEFORE the drop, never on it — an announcement that arrives simultaneously with the thing it announces is a collision. The build checks the bar arithmetic for exactly that, and the check had to be written carefully: floor division on a product of floats reports a downbeat as the bar before it, and rounding instead pushes a beat-3.5 shout into the bar after. Nudge, then floor. Loop seam 1.890 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2740955, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain); Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "deadhead": { "src": "music/deadhead.ogg", "ms": 96000, "loop": true, "bpm": 150, "key": "C#2 aeolian (INSTRUMENTAL; static riff under a lead that climbs)", "notes": "AN INSTRUMENTAL, AND THAT IS THE WHOLE DESIGN. A deadhead run is a truck driving with no load; this is a record with no vocal, the first in this catalogue. 150 bpm at a 75 feel, C# minor, trap metal. NO LYRICS IS A COMPOSITIONAL CONSTRAINT RATHER THAN AN OMISSION. Every other record here leans on a voice to carry identity — the hook is a phrase, the character is a delivery. Remove that and the guitar has to do both jobs, which means this record needs two things the catalogue learned SEPARATELY and had never needed together, and which directly contradict each other. I GET THE LOAD established that a trap hook works BECAUSE it does not develop: repetition is the mechanism, and that build refuses a hook whose consecutive statements differ by more than one note. JAKE BRAKE established the opposite for a lead: register, density and bend depth must climb across a solo or it is a scale with opinions, and that build refuses a flat arc. An instrumental needs BOTH AT ONCE, because a record that only repeats is a loop and a record that only develops has nothing to develop AGAINST. So this is the only track here where those two gates coexist. Measured: the chug figure runs 2 patterns across 54 bars and is 79% identical to its own neighbour, while the lead climbs octave 2 to 3, bend depth 1.0 to 3.4 semitones, and note density 2 to 4 subdivisions per beat. The riff gate earned its place immediately — the first cut alternated every fifth bar and measured 67% against a 70% floor. It was right: a figure that changes every five bars is not something an ear settles into, it is a two-bar phrase in disguise. THE GNARL IS THREE SEPARATE MECHANISMS, because “distorted” is not one thing and the build requires all three. PALM MUTES at high drive — short, dry, pitched percussion, the floor of the record, 660 of them. POWER CHORDS summed BEFORE the gain stage, so the amp generates the intermodulation between root, fifth and octave that IS the sound of a power chord rather than three clean notes added together, 714 of them. PINCH HARMONICS driven at the fourth partial — physically what a finger resting on a node does, not an EQ and not a pitch shift, 101 of them. Plus 163 bent notes across dives, scoops, rises and flutters, gated at a minimum of 40, because with no voice on the record the bar is the only expressive gesture left. THE BASS IS THE IN-HOUSE CHAIN tuned deep: `domebass` asymmetric saturation at asym 0.70 with a sub-lift shelf, plus a gliding second layer at asym 0.44. The asymmetry is load-bearing — at asym 0 the stage is plain tanh and generates no even harmonics, and those are what let a very low note be audible on a speaker that cannot reproduce its fundamental. THE ABSENCE OF A VOICE IS ASSERTED IN BOTH FILES. The composer refuses to build if vocal takes exist on disk and the mixer refuses to master, because a mixer copied from a neighbouring track would quietly place vocals if any happened to be there, and that would be a different song. Loop seam 0.108 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2623132, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "who_clipped_me": { "src": "music/who_clipped_me.ogg", "ms": 118652, "loop": true, "bpm": 89, "key": "D2 aeolian, DROP D (instrumental; wavefolded bass, phase-shift self-duet)", "notes": "STUDIED FROM YG — “Who Shot Me?” (Still Brazy, 2016). 89 bpm, West Coast, an album characterised by thick bass lines and a dark paranoid minimalism. Clipping a mirror is what a trucker calls a sideswipe, which carries the same double meaning the original title does. 89 BPM IS THE SLOWEST TEMPO IN THIS CATALOGUE and that shapes everything. CHROME DRIP at 100 was already too slow for a half-time feel to work; at 89 a bar runs 2.70 seconds and every single event is individually audible. There is nowhere for a weak sound to hide, so this record spends its effort on the QUALITY of four sounds rather than on the complexity of the arrangement. It is the second instrumental here — no vocal, asserted in both files. THE BASS IS THREE INSTRUMENTS AND ONE OF THEM WAS BUILT FOR THIS RECORD. SUB is a near-sine doing nothing but weight. DEEP is the drop bass this catalogue already uses — `domebass` asymmetric saturation with a sub-lift shelf, whose even harmonics let a very low note survive a small speaker. GNARLY is new, and it is a different MECHANISM rather than more of the same one: where saturation flattens a hot signal into a plateau and yields a tidy harmonic series, WAVEFOLDING sends the excursion back down, so the transfer curve is non-monotonic and the result is dense and inharmonic. A resonant band-pass parked at 300-900 Hz then sits that nastiness exactly where a guitar body would be rather than under it. AND THE NEW SOUND IS PROVED RATHER THAN CLAIMED. “I made a nastier bass” is exactly the kind of assertion that is never checked, so the bake is measured with `timbre.py` — the same tool that separates a fuzz-guitar record from an upright-bass one — and the two must land far apart or the build fails. Measured: centroid 78.6 Hz against 311.1, even/odd 6.44 against 0.00, harmonic-to-percussive 115.3 against 10.0, timbre distance 2.000 out of 2.000. The even/odd collapse and the low HPR are the wavefold doing what it says: the output is not a harmonic series at all. GNARLY is also the MOVING voice, carrying 96 notes against DEEP's 12 — asserted, because a new sample loaded into the old part is not a new instrument used. THE GUITAR IS DROP D, ECHOED, AND SELF-DUETED BY A NINETY-DEGREE PHASE SHIFT — WHICH IS NOT A DELAY, and that distinction is the whole reason to use it. A copy delayed by a few milliseconds is the same waveform arriving twice: summed, every frequency whose period divides that delay cancels, and you get comb filtering — a hollow notched version of one guitar. A copy shifted ninety degrees in PHASE, the Hilbert transform, has the SAME MAGNITUDE SPECTRUM and a completely different waveform, so it sums to something twice as wide with no notches anywhere: every frequency is decorrelated by the same QUARTER CYCLE rather than by a fixed TIME. To do that honestly the guitar has to exist as its own audio, so the composer renders TWO files — the bed, and the duet guitar ALONE — because a phase shift applied across a full mix decorrelates the drums too, and that is a stereo widener rather than a second guitarist. THREE PROPERTIES ARE ASSERTED, because a phase shift that silently degraded into a delay would sound almost right and be wrong for a reason nobody could hear their way to. The two channels must be nearly UNCORRELATED (measured 0.0000; a delay would be high). Their magnitude spectra must be nearly IDENTICAL (measured 0.0001 dB apart; a filter would not be). And the pair must SURVIVE A MONO FOLD (measured 0.1168 to 0.1172 RMS) — a widener that vanishes when summed to mono is a trick rather than an arrangement, and a great many listeners are on one speaker. A delayed copy fails the first and the third; a filter fails the second. The echo is locked to the dotted eighth at 506 ms and applied BEFORE the split, so both voices carry the same repeats. Trap hats throughout with ratchets on the drops. Loop seam 0.000 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 3004755, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "what_clipped_me": { "src": "music/what_clipped_me.ogg", "ms": 96603, "loop": true, "bpm": 240, "key": "D2 aeolian, DROP D (grindcore; 89 bpm for 30s then 240 and four blast types)", "notes": "THE THIRD RECORD ON ONE RIFF, AND THE ONE THAT DESTROYS IT. WHO CLIPPED ME is a beat; WHO CLIPPED ME II made it a song with a melody and a breakdown. This takes the same drop-D material and rebuilds it as GRINDCORE — the brief being that it may start slow, but by thirty to forty-five seconds it should rival Napalm Death for brutality. BRUTALITY IS A MEASUREMENT HERE RATHER THAN AN ADJECTIVE, which is the only way a brief like that can be met instead of gestured at. Three numbers, all asserted at build time. WHEN: the grind must begin inside the 30-45 second window that was actually specified. The record carries a real tempo map — 89 bpm through a two-bar accelerando to 240 — and the build computes the wall-clock second the blast starts and refuses anything outside the window. Measured: 30.6 seconds. The ramp is geometric in the PERIOD rather than linear in the BPM, the same reasoning CHROMEBEARD established: tempo is perceived as a ratio, so a straight line in the log domain is what accelerates evenly to the ear. HOW FAST: a blast beat is not merely fast drumming. It is the kick and snare ALTERNATING at eighth notes or finer so the kit fires one onset per subdivision, and the build requires at least ten percussion onsets per second sustained across the grind. Measured: 16.1. For scale, the densest previous record in this catalogue managed 20.8 events per second across EVERY instrument; this one runs 47.6. HOW LONG: grindcore is short and total, so more than half the record must sit at grind tempo or it is a slow song with a fast bit in it. Measured: 56%. TREMOLO PICKING IS THE OTHER HALF OF THE SOUND, and it is a different thing from a fast riff: one note repeated as fast as the hand moves, so pitch stops being rhythm and becomes TEXTURE. The guitar plays 32nd notes on a single degree for whole bars — at 240 bpm that is 32 notes per second per guitar, doubled — across 54 bars. Critically it is NOT palm-muted, and the build asserts that: a muted tremolo is a chug, and the wash is the entire point. The 808 stops being a bass LINE and becomes a pedal on the root, because at this tempo a moving bass is mud. THE GROWLS ARE NOT TUNED AND CANNOT BE. A death growl is a false-fold sound — the ventricular folds vibrate beneath the true folds, roughly an octave down and CHAOTICALLY, which is why it reads as inhuman rather than merely angry and why a pitch tracker finds no stable fundamental in it at all. A retuner run over one finds the tracker's noise floor and bends that, turning a roar into a warble. So they are placed dry with a 62 ms slap locked to a sixteenth, and the mixer imports no autotune — asserted by absence, the same rule BOOST GAUGE's drop samples established for the same physical reason. AND NO VOICE LANDS IN THE SLOW OPENING, also asserted. The first thirty seconds are the old song being recognisable before it is taken apart; the voice arrives WITH the violence rather than announcing it. First growl at 32.6 seconds. KEPT FROM THE FIRST TWO RECORDS: drop D, the wavefolded GNARLY bass, and the riff's own shape at the opening, so the listener knows what is being destroyed. Loop seam 0.767 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2266177, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "son_of_diesel_money": { "src": "music/son_of_diesel_money.ogg", "ms": 94648, "loop": true, "bpm": 142, "key": "D2 aeolian, DROP D (the melody IS bent power chords; no single notes)", "notes": "THE SEQUEL TO DIESEL MONEY. It keeps that record's 142 bpm and its trap grammar and replaces everything that made the sound: drop D instead of F# minor, a wavefolded gnarly bass instead of a sliding 808 line, scratching, dirty drums — and a melody carried entirely by BENT POWER CHORDS, echoed and self-dueted. The three one-liners are delivered by the DOWN FOR ROBOTICS character, a wrecked gravel-bottomed narrator who does not shout because the beat is already doing it. THE MELODY BEING POWER CHORDS IS THE RECORD, AND IT IS BACKWARDS ON PURPOSE. A power chord is HARMONY — root, fifth, octave, no third — and it is normally what a melody sits on top of. Handing it the tune means the melodic motion has to come from two places that are usually decoration: WHERE THE CHORD MOVES, and HOW FAR EACH ONE BENDS. A scoop into a chord is a different melodic gesture from landing on it square, and with no single notes available that contrast IS the phrasing. There is no lead line anywhere on this record and the build REFUSES one — the moment a single-note melody appears the power chords revert to accompaniment and the whole idea has quietly become a normal arrangement. Measured: 93 power chords, 74% of them bent across three shapes, ZERO single notes. WHY IT WORKS RATHER THAN JUST BEING A RESTRICTION: a bent power chord moves its root, fifth and octave TOGETHER, so the interval structure stays fixed while the whole stack slides. That is a sound a single note cannot make and a chord progression cannot make either — closer to a pedal steel than to a guitar, and at drop-D pitch it is enormous. THE BASS DOES TWO JOBS AND THE BRIEF SEPARATED THEM PRECISELY. GNARLY carries the rhythm — a wavefolded, band-passed 808 whose upper spectrum is dense and inharmonic rather than a tidy harmonic series. DEEP arrives in a few places to pick it up: smooth asymmetric saturation with a sub-lift shelf, and only six of them in the whole record, so they are EVENTS rather than a second bass part. The build asserts that ratio (216 against 6) because a drop hit that appears every bar is not a drop hit. THE MELODY IS SELF-DUETED BY A NINETY-DEGREE PHASE SHIFT and is rendered to its own file so the shift lands on power chords and nothing else — a Hilbert quadrature applied across a full mix decorrelates the drums too, which is a stereo widener rather than a second guitarist. Three properties asserted: correlation 0.0000, magnitude spectra 0.0002 dB apart, and it survives a mono fold. A delayed copy fails the first and the third; a filter fails the second. THE ONE-LINERS ARE BENT, NOT RESYNTHESISED, AND ONLY BARELY. This project has two vocal tools and choosing wrongly has cost whole rounds: one rebuilds a voice from an LPC fit of the tract (right for sustained singing), the other bends the actual recording (right when the character IS the delivery). This character is nothing but delivery — gravel, space and certainty — so melodies are derived from each take's own measured f0 and the correction only nudges it into the key. Measured: home notes G2, A#2 and A2, every one within 0.3 semitones of where that take already sat; tuning 18-24 cents. Anything more would sand off what makes him him. THREE LINES IN NINETY-FIVE SECONDS, minimum sixteen bars apart, asserted. A one-liner is a complete thought that lands once; three of them stacked into a verse is a paragraph. Loop seam 0.134 against a house gate of 12.0. NOTHING HERE WAS HEARD.", "bytes": 2674277, "credit": "Drums: AVL Drumkits “Black Pearl 4pc” by Glen MacArthur & Robin Gareus, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "far_horizon": { "src": "music/far_horizon.ogg", "ms": 120000, "loop": true, "bpm": 60, "key": "D dorian, free (idle: pads and slow strings, 0.8 events/s)", "notes": "The most ambient track in the batch: warm and halo pads, slow strings, no percussion of any kind, 0.80 events per second. The one rhythmic event is a sub swell at each phrase start, and its attack MEASURES 272 ms off the rendered low band — about ten cycles of its own 36.7 Hz fundamental, so it arrives as weight rather than as a knock. That number was wrong twice before it was measured: the original claimed 20 ms and called it slower than one cycle, which is inverted, and a correction invented 60 ms. The envelope is not a composer constant at all. A duration gate here was also unfireable — a guard skipped exactly the events the later assert claimed to check, so its only possible counterexamples were filtered out before it ran. Part of the IDLE SHELF — a batch written together so a dozen tracks by different hands still sound like one thing. They share `idlekit.py`, which is the style definition expressed as gates rather than as adjectives: sparse, no foreground, long notes, no forbidden percussion, and an invisible loop. IDLEKIT GATES THE COMPOSITION RATHER THAN THE MASTER, and that took three failed attempts to learn. Transient metrics on finished audio measured the MASTERING CHAIN, not the music — every track here is limited to -12.6 LUFS under a -5 dBFS ceiling, which flattens exactly the differences being measured. One audio check survives: the loop seam, because a seam is a discontinuity rather than a level and a limiter cannot flatten it. Its idle limit is 2.0 against the house gate of 12.0, because a menu loops for minutes while somebody reads. The forbidden-role gate DERIVES the drum roles from the rendered events by reverse lookup. It used to take them as an argument from the caller, and adversarial review proved that spoofable: a reviewer left one hand-written role declaration in place while playing a snare, and the build passed while rendering 160 snare hits. A gate whose evidence comes from the thing being gated is not a gate. NOTHING HERE WAS HEARD.", "bytes": 2379567, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "ocarina_flats": { "src": "music/ocarina_flats.ogg", "ms": 114286, "loop": true, "bpm": 88, "key": "mixolydian, 88 bpm (idle: wind lead over plucked, original singable melody)", "notes": "A wind lead over plucked accompaniment, mixolydian rather than plain major, with harmony that sidesteps more often than it resolves. The melody is ORIGINAL and is asserted SINGABLE the way this catalogue defines it — span within nine semitones and at least 60% of intervals a tone or less, measured in SEMITONES rather than in scale degrees, because a line written in degrees can look stepwise on paper and measure as an arpeggio. The calmest thing in the batch by onset rate. The grammar is borrowed; nothing is transcribed. Part of the IDLE SHELF — a batch written together so a dozen tracks by different hands still sound like one thing. They share `idlekit.py`, which is the style definition expressed as gates rather than as adjectives: sparse, no foreground, long notes, no forbidden percussion, and an invisible loop. IDLEKIT GATES THE COMPOSITION RATHER THAN THE MASTER, and that took three failed attempts to learn. Transient metrics on finished audio measured the MASTERING CHAIN, not the music — every track here is limited to -12.6 LUFS under a -5 dBFS ceiling, which flattens exactly the differences being measured. One audio check survives: the loop seam, because a seam is a discontinuity rather than a level and a limiter cannot flatten it. Its idle limit is 2.0 against the house gate of 12.0, because a menu loops for minutes while somebody reads. The forbidden-role gate DERIVES the drum roles from the rendered events by reverse lookup. It used to take them as an argument from the caller, and adversarial review proved that spoofable: a reviewer left one hand-written role declaration in place while playing a snare, and the build passed while rendering 160 snare hits. A gate whose evidence comes from the thing being gated is not a gate. NOTHING HERE WAS HEARD.", "bytes": 2015314, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "cantina_dust": { "src": "music/cantina_dust.ogg", "ms": 110000, "loop": true, "bpm": 100, "key": "phrygian, 100 bpm (idle: arpeggiated Spanish guitar, flat second asserted)", "notes": "Phrygian arpeggiated finger-style acoustic guitar, the warmest track in the batch. The flat second is the entire flavour and is asserted present. Deliberate small timing offsets keep it off the grid. The trap this had to avoid: flamenco is normally percussive and fast, which is the opposite of the brief — so it takes the harmony and the arpeggio texture and leaves the rasgueado behind. Part of the IDLE SHELF — a batch written together so a dozen tracks by different hands still sound like one thing. They share `idlekit.py`, which is the style definition expressed as gates rather than as adjectives: sparse, no foreground, long notes, no forbidden percussion, and an invisible loop. IDLEKIT GATES THE COMPOSITION RATHER THAN THE MASTER, and that took three failed attempts to learn. Transient metrics on finished audio measured the MASTERING CHAIN, not the music — every track here is limited to -12.6 LUFS under a -5 dBFS ceiling, which flattens exactly the differences being measured. One audio check survives: the loop seam, because a seam is a discontinuity rather than a level and a limiter cannot flatten it. Its idle limit is 2.0 against the house gate of 12.0, because a menu loops for minutes while somebody reads. The forbidden-role gate DERIVES the drum roles from the rendered events by reverse lookup. It used to take them as an argument from the caller, and adversarial review proved that spoofable: a reviewer left one hand-written role declaration in place while playing a snare, and the build passed while rendering 160 snare hits. A gate whose evidence comes from the thing being gated is not a gate. NOTHING HERE WAS HEARD.", "bytes": 2246398, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "paper_lantern": { "src": "music/paper_lantern.ogg", "shelved": "2026-09-01, operator: pulled from the in-game rotation pending a rework. Kept as a record so putting it back is one line in shop.alts, and so the gate can tell a deliberate shelving from an accidental orphan.", "ms": 116364, "loop": true, "bpm": 70, "key": "D hirajoshi, 66 bpm (idle: self-dueted bent guitar over a drone)", "notes": "Plucked sustained voices in a pentatonic idiom, the sober cousin of this catalogue's DRUNK STRINGS sound. No note outside the chosen five-note set appears anywhere, asserted. Phrases breathe with real silence between them, and slides between degrees are the ornament that makes the idiom read as itself rather than as generic pentatonic wallpaper. Its encodes were written to the record's own directory rather than the shared one because five agents were writing concurrently — correct at the time, and the reason is recorded rather than tidied away. Part of the IDLE SHELF — a batch written together so a dozen tracks by different hands still sound like one thing. They share `idlekit.py`, which is the style definition expressed as gates rather than as adjectives: sparse, no foreground, long notes, no forbidden percussion, and an invisible loop. IDLEKIT GATES THE COMPOSITION RATHER THAN THE MASTER, and that took three failed attempts to learn. Transient metrics on finished audio measured the MASTERING CHAIN, not the music — every track here is limited to -12.6 LUFS under a -5 dBFS ceiling, which flattens exactly the differences being measured. One audio check survives: the loop seam, because a seam is a discontinuity rather than a level and a limiter cannot flatten it. Its idle limit is 2.0 against the house gate of 12.0, because a menu loops for minutes while somebody reads. The forbidden-role gate DERIVES the drum roles from the rendered events by reverse lookup. It used to take them as an argument from the caller, and adversarial review proved that spoofable: a reviewer left one hand-written role declaration in place while playing a snare, and the build passed while rendering 160 snare hits. A gate whose evidence comes from the thing being gated is not a gate. NOTHING HERE WAS HEARD.", "bytes": 2178669, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "something_coming": { "src": "music/something_coming.ogg", "ms": 116757, "loop": true, "bpm": 78, "key": "D aeolian (CAUTIOUS: unresolved drone, off-grid events, no sting)", "notes": "The one CAUTIOUS track: something is out there and the player has not seen it. A low sustained drone with small irregular events above it that arrive off-grid, so the ear cannot predict them. It never lands on a plain root-position tonic — the tension is in what does not happen. idlekit's cautious mood relaxes exactly two limits and NOT the two that matter: it may not be denser, and it may not park energy where the game's own warning cues live, because a warning the player cannot hear is a bug. It also passes the stricter CALM limits. Part of the IDLE SHELF — a batch written together so a dozen tracks by different hands still sound like one thing. They share `idlekit.py`, which is the style definition expressed as gates rather than as adjectives: sparse, no foreground, long notes, no forbidden percussion, and an invisible loop. IDLEKIT GATES THE COMPOSITION RATHER THAN THE MASTER, and that took three failed attempts to learn. Transient metrics on finished audio measured the MASTERING CHAIN, not the music — every track here is limited to -12.6 LUFS under a -5 dBFS ceiling, which flattens exactly the differences being measured. One audio check survives: the loop seam, because a seam is a discontinuity rather than a level and a limiter cannot flatten it. Its idle limit is 2.0 against the house gate of 12.0, because a menu loops for minutes while somebody reads. The forbidden-role gate DERIVES the drum roles from the rendered events by reverse lookup. It used to take them as an argument from the caller, and adversarial review proved that spoofable: a reviewer left one hand-written role declaration in place while playing a snare, and the build passed while rendering 160 snare hits. A gate whose evidence comes from the thing being gated is not a gate. NOTHING HERE WAS HEARD.", "bytes": 2227465, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "dust_and_diesel": { "src": "music/dust_and_diesel.ogg", "ms": 129730, "loop": true, "bpm": 74, "key": "D dorian, 74 bpm (idle: lo-fi menu bed; kick and hats only)", "notes": "74 bpm, D dorian, 40 bars. Electric piano rootless voicings, vibraphone figure, acoustic bass, warm pad, and a light kit playing KICK AND HATS ONLY — the backbeat on beat 3 is a bass note rather than a drum, because idlekit forbids the snare outright and a snare is not made acceptable by being quiet. The arrangement is a closed Gray path over ten states: exactly one of five switches changes per repeat INCLUDING the wrap, so the loop point is the smallest change in the record rather than the largest. The hard part of a four-bar loop is making it bear ten minutes, and the answer is to vary one thing at a time and never two. Part of the IDLE SHELF — a batch written together so a dozen tracks by different hands still sound like one thing. They share `idlekit.py`, which is the style definition expressed as gates rather than as adjectives: sparse, no foreground, long notes, no forbidden percussion, and an invisible loop. IDLEKIT GATES THE COMPOSITION RATHER THAN THE MASTER, and that took three failed attempts to learn. Transient metrics on finished audio measured the MASTERING CHAIN, not the music — every track here is limited to -12.6 LUFS under a -5 dBFS ceiling, which flattens exactly the differences being measured. One audio check survives: the loop seam, because a seam is a discontinuity rather than a level and a limiter cannot flatten it. Its idle limit is 2.0 against the house gate of 12.0, because a menu loops for minutes while somebody reads. The forbidden-role gate DERIVES the drum roles from the rendered events by reverse lookup. It used to take them as an argument from the caller, and adversarial review proved that spoofable: a reviewer left one hand-written role declaration in place while playing a snare, and the build passed while rendering 160 snare hits. A gate whose evidence comes from the thing being gated is not a gate. NOTHING HERE WAS HEARD.", "bytes": 2537659, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "cold_start": { "src": "music/cold_start.ogg", "ms": 124000, "loop": true, "bpm": 0, "key": "D1 minor pentatonic (no tempo — a free-running patch)", "notes": "BPM IS ZERO BECAUSE THERE IS NO TEMPO. Every other record in this catalogue is grid-locked: events land on subdivisions of a clock and the loop repeats exactly. A patched semi-modular does the opposite — one LFO on the filter, one on the pitch, one on the amplitude, one on the gate, none of them synced to a clock or to each other, so the combined state of the patch never returns to a configuration it has already been in. That, not the timbre, is what the reference actually sounds like, and because it is a property of the composition it is gated rather than hoped for. The four modulator periods (16.5 / 25.2 / 34.0 / 49.8 s) were found by SEARCH, not chosen by eye. Hand-picked awkward-looking numbers put two of them at 8/3 to within 0.3% and the patch relocked every 93 s inside a 124 s record — eyeballing irrationality does not work. The gate that caught it was itself rewritten first: it had asked whether each ratio was far from every p/q with q < 9, which is very nearly unsatisfiable, since fractions with q <= 8 sit about 0.018 apart and almost every real number is within 0.02 of one. Even the golden ratio fails that test. The question worth asking is not whether a ratio is irrational but whether the pair RELOCKS INSIDE THE RECORD, so the gate now finds the simplest fraction each ratio is genuinely near and requires the resulting cycle to outlast the track. No pair of these four is within 1% of any p/q with q <= 12 at all. Onsets are placed by the modulators rather than on a subdivision, and the build measures how well they fit ANY common grid and refuses a good fit — 21% here, where a sequence would read near 100%. The oscillators drift as an Ornstein-Uhlenbeck process, mean-reverting rather than a free random walk whose spread grows without bound and would eventually be simply out of tune. The sub is WAVEFOLDED rather than saturated, because folding sends the excursion back down instead of flattening it and makes dense inharmonic content rather than a warm edge. One gate lives in the mixer and is specific to this record: a limiter is gain reduction keyed on level, which is exactly a device for flattening slow amplitude modulation — and one of the four modulators IS the amplitude. So the record can pass every composition gate, master to a perfect number, and arrive with its defining property removed. The mastering chain therefore measures the depth of the 49.8 s envelope component before and after and refuses to ship under 55% retention; it came through at 98.7%, and the measurement was proved capable of failing by flattening the bed with a slow AGC, which read 21.5%. NOTHING HERE WAS HEARD.", "bytes": 2193890, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "radio_kept": { "src": "music/radio_kept.ogg", "ms": 118000, "loop": true, "bpm": 0, "key": "A1 minor pentatonic (no tempo — a patch, haunted by three in-game measures)", "notes": "A minor pentatonic, 118 s. Two slow breaths across the record — the amplitude modulator is EXACTLY 2 cycles in the loop, against TAPE ROT's three, which is a real difference in pace rather than different numbers. Loop-alignment stops being optional once the modulation is deep: a record LOOPS, so an LFO that ends at a different phase than it started puts a level jump at the loop point, and the other three periods were re-searched around the fixed amp period so the four still never return to the same relative phase. It remembers nasty_on_delayed_arrival, deadhead and who_clipped_me. THIS RECORD FORGETS BY DISTANCE: a small speaker's band, then discrete echoes with feedback — a signal that travelled, bounced and arrived thin, and you can hear the repeats. The echoes are applied BEFORE the filter, because filtering first would let each repeat re-introduce top end the filter had just removed and the tail would get brighter as it decayed, which is the opposite of what distance does. Its bed is a drone pair an OCTAVE apart rather than a fifth: an octave beats plainly, and plain is right when the record's interest is meant to arrive from outside it. Part of the HAUNTED PATCHES: a free-running semi-modular patch — four mutually incommensurate modulators, no grid, bounded Ornstein-Uhlenbeck drift on the oscillators — with three memories of the game's own music surfacing through it. THE MEMORIES ARE ONE MEASURE EACH, drawn at random with a fixed seed from the live arena manifest's ACTIVE pools: the records a player has actually been hearing during a fight. Exactly one measure at the SOURCE's own tempo, derived from its bpm rather than a round number of seconds, so what surfaces is a musical unit and not a slice; the offset inside each source is drawn from its strongest riff windows, random but not random enough to land on an empty bar. Each plays at A QUARTER OF ITS SOURCE'S OWN LEVEL, and the reference is the whole point. The first version of this pair gated the fragments against THE BED instead, which sounds equivalent and is not: measured afterwards, that had them 4 to 7 times quieter than a quarter of their source, so the memories were nearly inaudible while the bed was too loud — wrong in both directions at once. The bed is a moving, arbitrary reference; the source recording is a fixed one, and 'a quarter as loud as that track normally is' is something a listener can actually judge. It is measured over the fragment's BODY and not its tail, because the tail is a decay by construction and averaging it in would silently push the body louder to compensate. Where each memory lands is chosen from the bed's OWN ENVELOPE rather than by eye — once the record breathed deeply, a time picked by hand put a fragment squarely in a trough where it stood at 3.1x the music under it. A ghost must still be materially darker than its source (under 0.80 of its spectral centroid), because distance and dullness are the same physical fact and a fragment with its top end intact reads as in-the-room at any level; that ceiling was RELAXED from 0.70 after listening, since at 0.70 the fragments were dulled past recognition and the brief is that a memory SURFACES — you are meant to half-know what it was. PUBLISHED AT -18 LUFS, 5.4 LU below the rest of the shelf. That is a listening decision and not a measurement, which is worth saying plainly: the first attempt to justify it went looking for physics to support a conclusion already chosen — the theory that a sustained drone SITS at its integrated loudness while a percussive record only touches it on peaks. Measured across the catalogue that is false and backwards (ambient records spend 45% of their length within 3 dB of their loud level against the active records' 89%, because the active ones are limited walls that live at the ceiling). What the warning about single tones is really about is PROMINENCE rather than loudness — the same thing that made PAPER LANTERN jarring while every gate passed — so the sustained voice in each bed was cut to roughly a third and lengthened, to read as part of the air instead of a line played at you. Publishing these also exposed a defect in the shared publisher: `render.publish()` forwards its loudness target to the mp3 encoder ONLY, while the ogg is PEAK-normalised, so a published ogg's loudness is a consequence of the master's crest rather than anything anyone set. Measured here at -12.6 (ogg) against -18.2 (mp3), a 5.6 dB split inside one release; the rest of the catalogue matches only by coincidence. Both encodes are now made together at one level and asserted to agree. NOTHING HERE WAS HEARD.", "bytes": 2280969, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "tape_rot": { "src": "music/tape_rot.ogg", "ms": 132000, "loop": true, "bpm": 0, "key": "E1 minor pentatonic (no tempo — a patch, sagging; three in-game measures)", "notes": "E minor pentatonic, 132 s, three breaths. It remembers chromebeard, who_clipped_me and nasty_on_delayed_arrival. THIS RECORD FORGETS BY DECAY rather than by distance, and that is what makes the two records two: no echoes at all, but an unreliable timebase — wow and flutter at two incommensurate rates so the wobble never becomes a pulse the ear can lock to, smeared into a long dull tail. Nothing repeats; it just sags. Its oxide floor FOLLOWS the amplitude LFO, which a real transport does anyway, and that is not decoration: a constant floor was tried first to fix a loudness shortfall and swamped the amp modulator, dropping its depth to 0.017 against 0.16 on its sibling, whereupon the mastering gate read the resulting ratio as the limiter pumping when it was really a ratio between two near-zero numbers. The loudness shortfall itself was CREST and not ceiling — lowering the peak ceiling to buy level made it measurably worse. Part of the HAUNTED PATCHES: a free-running semi-modular patch — four mutually incommensurate modulators, no grid, bounded Ornstein-Uhlenbeck drift on the oscillators — with three memories of the game's own music surfacing through it. THE MEMORIES ARE ONE MEASURE EACH, drawn at random with a fixed seed from the live arena manifest's ACTIVE pools: the records a player has actually been hearing during a fight. Exactly one measure at the SOURCE's own tempo, derived from its bpm rather than a round number of seconds, so what surfaces is a musical unit and not a slice; the offset inside each source is drawn from its strongest riff windows, random but not random enough to land on an empty bar. Each plays at A QUARTER OF ITS SOURCE'S OWN LEVEL, and the reference is the whole point. The first version of this pair gated the fragments against THE BED instead, which sounds equivalent and is not: measured afterwards, that had them 4 to 7 times quieter than a quarter of their source, so the memories were nearly inaudible while the bed was too loud — wrong in both directions at once. The bed is a moving, arbitrary reference; the source recording is a fixed one, and 'a quarter as loud as that track normally is' is something a listener can actually judge. It is measured over the fragment's BODY and not its tail, because the tail is a decay by construction and averaging it in would silently push the body louder to compensate. Where each memory lands is chosen from the bed's OWN ENVELOPE rather than by eye — once the record breathed deeply, a time picked by hand put a fragment squarely in a trough where it stood at 3.1x the music under it. A ghost must still be materially darker than its source (under 0.80 of its spectral centroid), because distance and dullness are the same physical fact and a fragment with its top end intact reads as in-the-room at any level; that ceiling was RELAXED from 0.70 after listening, since at 0.70 the fragments were dulled past recognition and the brief is that a memory SURFACES — you are meant to half-know what it was. PUBLISHED AT -18 LUFS, 5.4 LU below the rest of the shelf. That is a listening decision and not a measurement, which is worth saying plainly: the first attempt to justify it went looking for physics to support a conclusion already chosen — the theory that a sustained drone SITS at its integrated loudness while a percussive record only touches it on peaks. Measured across the catalogue that is false and backwards (ambient records spend 45% of their length within 3 dB of their loud level against the active records' 89%, because the active ones are limited walls that live at the ceiling). What the warning about single tones is really about is PROMINENCE rather than loudness — the same thing that made PAPER LANTERN jarring while every gate passed — so the sustained voice in each bed was cut to roughly a third and lengthened, to read as part of the air instead of a line played at you. Publishing these also exposed a defect in the shared publisher: `render.publish()` forwards its loudness target to the mp3 encoder ONLY, while the ogg is PEAK-normalised, so a published ogg's loudness is a consequence of the master's crest rather than anything anyone set. Measured here at -12.6 (ogg) against -18.2 (mp3), a 5.6 dB split inside one release; the rest of the catalogue matches only by coincidence. Both encodes are now made together at one level and asserted to agree. NOTHING HERE WAS HEARD.", "bytes": 2543725, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "greenhouse": { "src": "music/greenhouse.ogg", "ms": 92055, "loop": true, "bpm": 146, "key": "Ab2 MAJOR, 146 bpm (chopped break + trap hats; hook on bent self-dueted guitar)", "notes": "146 bpm, A-FLAT MAJOR, and the key is the study. Every other record in this catalogue is minor — aeolian, phrygian, hirajoshi, minor pentatonic, drop-D — because the game is a wasteland and minor is what menace sounds like. The reference is major, bouncy and pleased with itself, and no instrumentation choice substitutes for that, so this one goes to A-flat major and the gate asserts it BOTH ways: the major third must be PRESENT, because a record can be 'not minor' by dodging thirds entirely and that is neither, and the minor third must never appear. Nothing outside the scale does either. THE HOOK IS BENT GUITAR, SELF-DUETED. The reference's hook is a plucked mallet — short, dry, no sustain — and a bent guitar note is its opposite: it arrives, it MOVES, and it rings. Giving the same bouncy figure to it keeps the bounce and drags the timbre back into the game's world. It is ECHOED FIRST and dueted SECOND, and that order is not cosmetic: the quadrature then treats the repeats as part of the voice being doubled, where duetting first and echoing each side gives two independent delay lines drifting into a chorus. The echo is a dotted eighth (308 ms at this tempo) with damped taps, because a real repeat loses top end each time round, and it is PROVED PRESENT by autocorrelation at that lag (0.547 against a -0.013 neighbourhood) — a delay whose feedback silently went to zero looks identical in every log and leaves the hook dry. The duet is a Hilbert quadrature rather than a delayed copy: same magnitude spectrum, decorrelated waveform, so hard-panning it is twice as wide with none of the comb filtering a delay makes. Measured at correlation 0.0000, spectra 0.0002 dB apart, and the mono fold gets LOUDER not quieter, which is the check that catches a duet cancelling itself on a phone. THE DRUMS ARE A CHOPPED BREAK AND NOT A LOOP, which was the owner's plainest instruction — no repetitive stuff — so it is measured rather than intended. This file renders its own two-bar performance on a bop kit, cuts it on its own onsets into 39 slices, and re-sequences them; every bar gets a new pattern AND new slice choices, so two bars sharing a pattern still differ in the audio that plays it. Each bar reduces to a signature and the gate refuses any two consecutive bars that match, and any record under 90% distinct. Measured: 54 played bars, 54 distinct, 100%, zero consecutive repeats — where a four-bar loop run fourteen times would score 7%. A chopped break sounds unlike sequenced drums because the slices carry the room and the decaying tail of whatever was struck before the knife came down. Trap hats run over the top: the break is the groove, the hats are the grid it fights. 18 scratches, and TWO vocal takes as briefed — on a bright record that is not sparseness, it is what leaves the hook room to be the thing you remember. NOTHING HERE WAS HEARD.", "bytes": 2496496, "credit": "Drums: AVL Drumkits “Blonde Bop” by Glen MacArthur, licensed CC-BY-SA 3.0; Classic-808 — Hollow Sun (samples recorded by Steve Howell); Hydrogen kit by Artemiy Pavlov (Public Domain)", "license": "CC-BY-SA-3.0" }, "chrome_dome": { "src": "music/chrome_dome.ogg", "ms": 84000, "loop": true, "bpm": 0, "key": "C2 aeolian (no tempo — synthwave title; fan chopper, self-trio'd lead)", "notes": "84 s, C minor. THE TITLE RECORD, and a title record has one job the others do not: it has to ARRIVE. So it is built as a rise — the register opens from the bottom, voices accumulate, and the fan cuts across the top in four bursts at 13 Hz rather than running continuously. Its pads are an octave-and-third pair detuning against each other, its lead is a bent guitar figure trio'd hard left, centre and right, and the one vocal line ('Truck be chromed. Chill be domed.') arrives at 46 s in a low amused chest voice, itself trio'd. Published at -13.5 LUFS rather than the shelf's -12.6: a title screen is the first thing a player hears and it should not be the loudest. Part of the NEON PAIR, a title record and an idle one built together from the same study: '2059 INSTINCT OVER LOGIC // Synthwave, Downtempo, Cyberpunk' — a PLAYLIST rather than a song, so there is no hook to match and the study is a GENRE. What a genre study has to get right instead is REGISTER, and this one's is specific: wide, nocturnal, analog, and never clean. Per the brief nothing here is a pure tone — everything is driven, noisy, echoed, reverbed and self-harmonised by phase shift, and noise is used as an instrument rather than cleaned out of the way. Both are free-running patches underneath: four mutually incommensurate modulator periods found by search rather than by eye, nothing quantised, bounded Ornstein-Uhlenbeck drift on the oscillators, and the amplitude modulator dividing the loop exactly so the loop point is not a level jump. THE FAN is the new tool — 'chopping things up like talking through a fan' is amplitude occlusion at a rate the ear hears as flutter rather than pitch, roughly 6-30 Hz: below about 5 it is a tremolo you can count, above about 40 the sidebands fuse and it becomes timbre. The blade shape carries as much as the rate, because what makes a fan rather than a tremolo is that the blade OCCLUDES — most of the cycle open, a fast dark notch — so it is a raised cosine taken to a power, which sharpens the notch without the clicks a square wave would add, and its rate wobbles because a fan in a room does not hold tempo. It fades in and out in bursts like the scratches, by instruction: a constant chop is a texture you stop hearing inside ten seconds and the whole value is that it arrives and leaves. THE LEAD AND THE VOICE ARE SELF-TRIO'D, not dueted, and that needed a genuinely new piece of thinking. A Hilbert quadrature gives two voices with the same magnitude spectrum and no correlation, which is what makes a self-duet read as two players — but quadrature is a TWO-element basis and there is no third ninety degrees to take. Forty-five degrees correlates with both; a delay comb-filters; a pitch shift is a different instrument. The honest third voice is a random ALL-PASS: multiply the spectrum by a random phase curve and the magnitude is untouched by construction while the waveform decorrelates. The curve is smoothed rather than white, because fully random per-bin phase disperses transients into noise and a plucked note becomes a wash. ESTABLISHING THAT COST THREE WRONG HYPOTHESES and the record of it is in the tool: the first spectral check read 144 dB of difference between a signal and its own all-pass, which is impossible — it was averaging the dB ratio across sixty thousand EMPTY bins, comparing two near-zeros, the same mistake this project once made taking a median across silent frames. Gated to bins that have energy it read 1.08 dB, which was then blamed on truncation (padding made it worse), then on transform length (worse again), before the untruncated transform measured 0.000000 dB and settled it: the all-pass is exact and the CUT is the cost. So the transform is asserted at 0.05 dB where it is exact, and the trimming loss is asserted separately at 5% of energy — blending them into one loose threshold would let a real filter hide behind an expected truncation loss. Correlations are checked between all three pairs, and the all-pass seed is SEARCHED until they pass, which is ordinary (it is a free parameter); loosening the property would not be. The chop happens BEFORE the split so all three voices carry the same blade — chopping after would give three independent fans beating against each other, a tremolo choir rather than one player heard three ways. THE VOICES ARE LOW ON INSTRUCTION: 'don't make these high pitched', which is the opposite of the default a synth record invites. Both casts are chest voices, sultry and unhurried, sitting UNDER the pads rather than over them — the register good downtempo actually uses, where a hook works by holding back. One take each, used once. NOTHING HERE WAS HEARD.", "bytes": 1729470, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "night_haul": { "src": "music/night_haul.ogg", "ms": 120000, "loop": true, "bpm": 0, "key": "C1 aeolian (no tempo — downtempo cyberpunk idle; slower fan, one vocalise)", "notes": "120 s, C minor, and deliberately the opposite SHAPE of its sibling. CHROME DOME arrives; this one must never arrive at all — it plays under a menu or an empty desert while somebody reads, so idlekit gates it as calm (0.42 events/s against a 6.0 ceiling, nothing over amp 0.34, nothing under 0.18 s, no forbidden percussion) and its loop seam is 1.27 against the idle limit of 2.0, where the house gate is 12.0 — because a menu loops for minutes and a seam becomes a metronome long before a listener can name it. Same palette, opposite energy: the fan is slower (9 Hz) and rarer, the pads move in fourteen-second breaths, the sub arrives every twenty-six seconds, and the register never opens all the way. Its vocal is a wordless vocalise rather than a line — a shape, phase-shifted into a trio, drifting through once at 58 s. Published at -17.0 LUFS, 4.4 LU under the shelf, because ambient that plays for minutes should sit below music you chose to listen to. Part of the NEON PAIR, a title record and an idle one built together from the same study: '2059 INSTINCT OVER LOGIC // Synthwave, Downtempo, Cyberpunk' — a PLAYLIST rather than a song, so there is no hook to match and the study is a GENRE. What a genre study has to get right instead is REGISTER, and this one's is specific: wide, nocturnal, analog, and never clean. Per the brief nothing here is a pure tone — everything is driven, noisy, echoed, reverbed and self-harmonised by phase shift, and noise is used as an instrument rather than cleaned out of the way. Both are free-running patches underneath: four mutually incommensurate modulator periods found by search rather than by eye, nothing quantised, bounded Ornstein-Uhlenbeck drift on the oscillators, and the amplitude modulator dividing the loop exactly so the loop point is not a level jump. THE FAN is the new tool — 'chopping things up like talking through a fan' is amplitude occlusion at a rate the ear hears as flutter rather than pitch, roughly 6-30 Hz: below about 5 it is a tremolo you can count, above about 40 the sidebands fuse and it becomes timbre. The blade shape carries as much as the rate, because what makes a fan rather than a tremolo is that the blade OCCLUDES — most of the cycle open, a fast dark notch — so it is a raised cosine taken to a power, which sharpens the notch without the clicks a square wave would add, and its rate wobbles because a fan in a room does not hold tempo. It fades in and out in bursts like the scratches, by instruction: a constant chop is a texture you stop hearing inside ten seconds and the whole value is that it arrives and leaves. THE LEAD AND THE VOICE ARE SELF-TRIO'D, not dueted, and that needed a genuinely new piece of thinking. A Hilbert quadrature gives two voices with the same magnitude spectrum and no correlation, which is what makes a self-duet read as two players — but quadrature is a TWO-element basis and there is no third ninety degrees to take. Forty-five degrees correlates with both; a delay comb-filters; a pitch shift is a different instrument. The honest third voice is a random ALL-PASS: multiply the spectrum by a random phase curve and the magnitude is untouched by construction while the waveform decorrelates. The curve is smoothed rather than white, because fully random per-bin phase disperses transients into noise and a plucked note becomes a wash. ESTABLISHING THAT COST THREE WRONG HYPOTHESES and the record of it is in the tool: the first spectral check read 144 dB of difference between a signal and its own all-pass, which is impossible — it was averaging the dB ratio across sixty thousand EMPTY bins, comparing two near-zeros, the same mistake this project once made taking a median across silent frames. Gated to bins that have energy it read 1.08 dB, which was then blamed on truncation (padding made it worse), then on transform length (worse again), before the untruncated transform measured 0.000000 dB and settled it: the all-pass is exact and the CUT is the cost. So the transform is asserted at 0.05 dB where it is exact, and the trimming loss is asserted separately at 5% of energy — blending them into one loose threshold would let a real filter hide behind an expected truncation loss. Correlations are checked between all three pairs, and the all-pass seed is SEARCHED until they pass, which is ordinary (it is a free parameter); loosening the property would not be. The chop happens BEFORE the split so all three voices carry the same blade — chopping after would give three independent fans beating against each other, a tremolo choir rather than one player heard three ways. THE VOICES ARE LOW ON INSTRUCTION: 'don't make these high pitched', which is the opposite of the default a synth record invites. Both casts are chest voices, sultry and unhurried, sitting UNDER the pads rather than over them — the register good downtempo actually uses, where a hook works by holding back. One take each, used once. NOTHING HERE WAS HEARD.", "bytes": 2343007, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" }, "wig_tight": { "src": "music/wig_tight.ogg", "ms": 240000, "loop": true, "bpm": 0, "key": "C2 aeolian (no tempo — 7 sections, chopped and self-trio'd, instrumental)", "notes": "240 s, C minor. CHROME DOME's material and its one vocal, expanded into a composition. 'ENOUGH STRUCTURE TO QUALIFY AS A MUSICAL COMPOSITION' IS A CLAIM, so it is gated rather than asserted — an ambient record can drift for four minutes and call the drift an arrangement while every loudness and density number reads correct. Two assertions make the difference measurable. FORM: no two ADJACENT sections carry the same set of voices, because the moment the instrumentation changes is the only boundary a listener can actually hear, and a record whose voice-set never changes has one section however many names its source code gives it. Seven sections here — open, stir, voice, chug, break, rise, dissolve — and every adjacent pair differs. ARC: the per-section event density must have a SINGLE INTERIOR maximum, so flat fails (a texture) and monotonic fails (a ramp that never resolves). Measured: 10, 169, 171, 378, 212, 415, 10. The breakdown at 212 genuinely DROPS below the 378 before it and the 415 peak after — it drops the chug entirely, which is the arrangement decision that gate exists to protect, because the tempting thing is to keep every voice once it has entered and the result is a record that only ever accumulates. THE VOICE IS STRETCHED IN TWO OPERATIONS, NOT ONE. Resampling stretches by reading slower, so duration and pitch move together — one physical operation, which is why a slowed tape sounds lower — and taken alone to 3.4x it would have dropped this take nineteen semitones into a subsonic growl: psychedelic, and no longer a voice. So the LENGTH comes from granular overlap-add, whose cost (a smeared attack) is exactly what this record wants, and the WEIGHT from a separate bounded three-semitone resample. 4.0 s becomes 16.2 s, four times over, three arrivals. Then fan, then echo, then trio — order again: chopping before the split puts ONE blade on all three voices, where chopping after gives three blades beating into a tremolo choir, and the echo sits between so the repeats are chopped material rather than chopped repeats. ONE OF THE THREE IS DISTORTED, WHICH BREAKS THE TRIO'S SPECTRAL CONTRACT ON PURPOSE. neonkit.trio asserts all three share a magnitude spectrum, since that is what makes a phase-shift duet read as one player doubled rather than as a filter; distortion adds harmonics, so it happens AFTER the trio is proved and earns its own two assertions instead — the driven voice must still be DECORRELATED from its siblings (or the trio has collapsed to two) and must measurably GAIN high-frequency energy (or the drive stage did nothing and the log would not have said so). That second one caught a real defect: normalising by PEAK before tanh put a low-crest reverbed signal almost entirely in the straight part of the curve, and drive 2.6 moved the high-frequency share from 0.0735 to 0.0774, which is nothing. Normalising by RMS puts the average sample where the curve actually bends: 0.0735 to 0.1387, 1.9x, with correlation still 0.214. THE DRONE gets the same treatment at lower intensity — chopped in five bursts at 11 Hz written by the COMPOSER as part of the arrangement rather than left to the mix, then tripled and drowned — and because it is the one voice present throughout, it carries the arrangement by LEVEL instead of by entering and leaving, swelling into the rise and withdrawing for the dissolve. That is what makes a constant voice read as reactive. Trap hats fade in across `stir` rather than arriving at full level; the palm-muted chug is the engine of `chug` and is the voice the breakdown takes away; the scratch solo is the breakdown's whole content, echoed. Published at -15.0 LUFS, 2.4 LU under the shelf: a four-minute ambient overlay is not a record you lean into. NOTHING HERE WAS HEARD.", "bytes": 5597007, "credit": "GM GS Bank (public-domain General MIDI) — unattributed; shipped inside the Csound documentation package (Public Domain)" } } }